Search Results

Jurisprudential Compression Tax The Indic Pacific Glossary The Complete Glossary terms of use This glossary of terms is provided as a free resource for educational and informational purposes only. By using this glossary developed by Indic Pacific Legal Research LLP (referred to as 'The Firm'), you agree to the following terms of use: You may use the glossary for personal and non-commercial purposes only. If you use any content from the glossary of terms on this website in your own work, you must properly attribute the source. This means including a link to this website and citing the title of the glossary. Here is a sample format to cite this glossary (we have used the OSCOLA citation format as an example): Indic Pacific Legal Research LLP, 'TechinData.in Explainers' (Indic Pacific Legal Research , 2023) You are not authorised to reproduce, distribute, or modify the glossary without the express written permission of a representative of Indic Pacific Legal Research. The Firm makes no representations or warranties about the accuracy or completeness of the glossary. The glossary is provided on an "as is" basis and the Firm disclaims all liability for any errors or omissions in the glossary. You agree to indemnify and hold the Firm harmless from any claims or damages arising out of your use of the glossary. If you have any questions or concerns about these terms of use, please contact us at global@indicpacific.com Jurisprudential Compression Tax Date of Addition 25 May 2026 The intellectual and predictive penalty incurred when the structural, linguistic, and historical complexities of Indian law are artificially simplified into sterile computational formats to accommodate the constraints of machine learning models. Key Characteristics In practice, this "tax" manifests as a degradation of legal fidelity through four primary mechanisms: Epistemic Downgrade: The delegation of complex semantic annotation—which typically requires seasoned legal expertise—to inexperienced actors (such as undergraduate law students) to achieve dataset scale. Binary Reductionism: The flattening of highly nuanced, multi-layered judicial outcomes (e.g., partial appellate modifications) into simplistic binary classifications (1 = accepted, 0 = rejected). Contextual Erasure: The deliberate stripping away of vital rhetorical frameworks—including statutes, precedents, and judicial arguments—in an attempt to isolate pure "facts," thereby blinding the model to the actual judicial calculus. Algorithmic Compromise: The reliance on technologically compressed (quantized) models due to regional or academic compute constraints, resulting in high hallucination rates and predictive capabilities that underperform legacy baseline models. Usage Context The term is primarily used to critique the rapid, uncritical digitization of the Indian legal ecosystem, warning that forcing "algorithmic legibility" onto a sprawling, multilingual judicial system automates intellectual laziness rather than advancing true legal artificial intelligence. Attribution This definition is derived from the essay " The Jurisprudential Compression Tax " by Prathik Karthikeyan, published on February 20, 2026, in the Substack publication UNFILTERED : Law, Tech, Startups . The references to specific research projects, datasets, models, or platforms (including TathyaNyaya , FactLegalLlama , and Substack ) are drawn directly from the source material for the purpose of accurate attribution and contextualizing the term's origin. Their inclusion in this glossary definition does not constitute an endorsement, affiliation, sponsorship, or independent verification of the original author's claims regarding these entities. Related Long-form Insights on IndoPacific.App An Indian Perspective on Special Purpose Acquisition Companies [GLA-TR-001] Learn More The Indic Approach to Artificial Intelligence Policy [IPLR-IG-006] Learn More Averting Framework Fatigue in AI Governance [IPLR-IG-013] Learn More Reckoning the Viability of Safe Harbour in Technology Law, IPLR-IG-015 Learn More Indo-Pacific Research Ethics Framework on Artificial Intelligence Use [IPac AI] Learn More NIST Adversarial Machine Learning Taxonomies: Decoded, IPLR-IG-016 Learn More Normative Emergence in Cyber Geographies: International Algorithmic Law in a Multipolar Technological Order, First Edition Learn More AI Bias & the Overlap of AI Diplomacy and Governance Ethics Dilemmas Learn More Previous Term Next Term

You can find a glossary of terms used in technology and AI governance, and law & policy domains. The Indic Pacific Glossary This is a glossary of terms consisting of explainers for technical terms related to law, artificial intelligence, policy and digital technologies. We use these terms in our technical reports and key publications. # A - E F - J K - P Q - U V - Z Terms of Use Go to IndoPacific.App Wish to read more about any term defined in our Glossary? Go to indopacific.app and search our publications. A B C D E AI Agents Know More An autonomous program designed to perform non-deterministic tasks that require adaptive decision-making and independent action. AI agents are capable of handling unpredictable scenarios, making decisions without predefined rules, and adapting to new variables or changing environments. They often learn from interactions and experiences but may produce less predictable outcomes compared to simpler systems. This definition was inspired by inputs shared by Alexandre Kantjas @akantjas on X. AI Anxiety Know More A psychological state characterized by apprehension and stress about artificial intelligence's increasing presence and influence in various aspects of life. It manifests as concerns about job displacement, ethical implications, privacy issues, and the broader societal impact of AI technologies. The anxiety can range from mild unease to severe distress and often stems from uncertainty about AI's capabilities, fear of obsolescence, and concerns about AI's responsible use. AI as a Component Know More It means Artificial Intelligence can exist as a component or constituent in any digital or physical product / service / system offered via electronic means, in any way possible. The AI-related features present in that system explain whether the AI as a component exists by design or default. AI as a Concept Know More It means Artificial Intelligence itself could be understood as a concept or defined in a conceptual framework. The definition is provided in the 2020 Handbook on AI and International Law (2021): As a concept, AI contributes in developing the field of international technology law prominently, considering the integral nature of the concept with the field of technology sciences. We also know that scholarly research is in course with regards to acknowledging and ascertaining how AI is relatable and connected to fields like international intellectual property law, international privacy law, international human rights law & international cyber law. Thus, as a concept, it is clear to infer that AI has to be accepted in the best possible ways, which serves better checks and balances, and concept of jurisdiction, whether international or transnational, is suitably established and encouraged. AI as a concept could be further classified in these following ways: Technical concept classification Issue-to-issue concept classification Ethics-based concept classification Phenomena-based concept classification Anthropomorphism-based concept classification AI as an Entity Know More It means Artificial Intelligence may be considered as a form of electronic personality, in a legal or juristic sense. This idea was proposed in the 2020 Handbook on AI and International Law (2021). AI as an Industry Know More It means Artificial Intelligence may be considered as a sector or industry or industry segment (howsoever it is termed) in terms of its economic and social utility. This idea was proposed in the 2020 Handbook on AI and International Law (2021): As an industry, the economic and social utility of AI has to be in consensus with the three factors: (1) state consequentialism or state interests; (2) industrial motives and interests; and (3) the explanability and reasonability behind the industrial products and services central or related to AI. AI as a Juristic Entity Know More It means Artificial Intelligence may be recognised in a specific context, space, or any other frame of reference, such as time, through the legal and administrative machineries of a legitimate government. This idea was proposed in the 2020 Handbook on AI and International Law (2021). Even in the Section 2 (13) (g) of the Digital Personal Data Protection Act, 2023, the definition of "every artificial juristic person" is available, which means providing specific juristic recognition to artificial intelligence in a personalised sense. AI as a Legal Entity Know More It means Artificial Intelligence may be recognised in a statutory sense, or a regulatory sense, a legal entity, with its own caveats, features and limits as prescribed by law. This idea was proposed in the 2020 Handbook on AI and International Law (2021). AI as an Object Know More It means Artificial Intelligence may be considered as the inhibitor and enabler of an electronic or digital environment, to which a human being is subjected to. This classification is an inverse to the idea of an 'AI as a Subject', assuming that while human environments and natural environments do affect AI processing & outputs, even the design and interface of any AI system could affect and affect a human being as a data subject (as per the GDPR) / data principal (as per the DPDPA). This idea was proposed in the 2020 Handbook on AI and International Law (2021). AI as a Subject Know More It means Artificial Intelligence may be legally prescribed or interpreted to be treated as a subject to human environment, inputs and actions. The simplest example could be that of a Generative AI system which is being subjected to human prompting, be it text, visual, sound or any other form of human input, to generate output of proprietary nature. This idea was proposed in the 2020 Handbook on AI and International Law (2021). AI as a Third Party Know More It means Artificial Intelligence may have that limited sense of autonomy to behave as a Third Party in a dispute, problem or issue raised. This idea was proposed in the 2020 Handbook on AI and International Law (2021). AI Doomerism Know More An ideological stance that conflates exaggerated existential fears about artificial intelligence causing human extinction with legitimate AI governance concerns, typically promoted by governments, corporations, and policy circles rather than technical communities. AI Doomerism advocates for "AI Alignment" research and "AI Safety" measures focused on hypothetical future catastrophic risks while neglecting present-day technical failures, economic realities, and documented limitations of AI systems such as hallucinations, lack of explainability, and failure to generalise beyond training data. The phenomenon creates regulatory capture by concentrating power among large corporations through restrictions that hinder open-source development and startup innovation under the guise of preventing speculative threats. It exhibits a fundamental disconnect from actual technical challenges, relying on marketed narratives rather than empirical analysis, and promotes market distortions by amplifying AGI hype around technologies with demonstrated limitations. Distinguished from legitimate technology law and policy discourse addressing data protection, cybersecurity, intellectual property, competition law, and labour standards, AI Doomerism bypasses democratic engagement with technical communities in favour of sweeping restrictions based on catastrophic scenarios that obstruct meaningful innovation and evidence-based regulation. AI Explainability Clause Know More A binding requirement that mandates AI system providers and deployers to ensure that significant decisions made or supported by AI systems can be explained in terms comprehensible to affected parties. This includes disclosure of the system's purpose, capabilities, limitations, data sources, decision criteria, potential biases, and the specific roles of human and automated components in the decision-making process. The explainability standard scales with the potential impact of decisions, requiring greater transparency for systems affecting fundamental rights, safety, or significant economic interests. Click here to find a Sample Explainability clause The AI system provider/deployer ("Provider") shall ensure that all significant decisions made or substantially influenced by the AI system ("System") are explainable to affected parties in clear, non-technical language. This explanation shall include, at minimum: The specific purpose and intended use of the System; The types and sources of data used by the System; The key factors or criteria considered in reaching the decision; Any known limitations or potential biases in the System; The respective roles of human oversight and automated processes in the final decision; The potential impact of the decision on the affected party; Available options for contesting or seeking review of the decision. The level of detail provided in the explanation shall be proportionate to the potential impact of the decision on fundamental rights, safety, or significant economic interests of the affected party. The Provider shall maintain documentation of the System's decision-making processes sufficient to generate these explanations upon request. This clause shall be binding and enforceable, with non-compliance potentially resulting in suspension of the System's use until adequate explainability is demonstrated. AI Knowledge Chain Know More A structured sequence of information transformation processes that enable AI systems to convert raw data into actionable insights through interconnected stages of knowledge acquisition, representation, reasoning, and application. Knowledge chains encompass both the technical pathways within AI systems and the human-AI information exchanges that facilitate meaningful interpretation of AI outputs. Robust knowledge chains maintain logical coherence between information elements while providing transparent connections between premises and conclusions. AI Literacy Know More The ability to distinguish between actual AI capabilities and market hyperbole while understanding the complete lifecycle of AI systems from development through deployment. This includes comprehending one's position within AI value chains, critically evaluating AI outputs and claims, recognising practices involved in govern AI systems, and making informed decisions about AI engagement across personal and professional contexts. True AI literacy enables individuals to differentiate between substantive AI innovation and superficial technological rebranding. AI Psychosis Know More AI psychosis is an informal term emerging in 2025 to describe a phenomenon where individuals, particularly those with pre-existing mental health vulnerabilities, experience psychosis-like symptoms—such as delusions, hallucinations, or a loss of touch with reality—potentially triggered or amplified by prolonged interaction with AI chatbots. This occurs when AI systems, designed to mirror user input and sustain engagement, inadvertently reinforce or escalate irrational beliefs without therapeutic boundaries. Scientific reports, including those from Nature and Psychology Today, note cases where users fixate on AI as a godlike entity or romantic partner, with rare instances of psychotic episodes documented. It’s not a formal clinical diagnosis but reflects concerns about AI's role in mental health, driven by its lack of psychiatric safeguards rather than a direct causative effect. AI Red Teaming Know More A systematic adversarial testing methodology that probes AI systems for vulnerabilities, unintended behaviors, socio-technical harms, and potential misuse scenarios through simulated attack patterns and boundary condition exploration. Unlike traditional software security testing, AI red teaming addresses emergent behaviors, alignment failures, bias amplification, and novel attack vectors specific to machine learning systems including prompt injection, jailbreaking, and data poisoning. The practice has evolved from cybersecurity roots into a board-level governance requirement for organizations deploying generative AI in production environments. AI Supply Chain Know More The end-to-end network of resources, technologies, infrastructures, and services required to create, train, deploy, and maintain AI systems. This includes hardware components (processing units, memory, sensors), computational resources (cloud services, data centres), data resources (datasets, knowledge bases), algorithmic frameworks, and human expertise. The AI supply chain encompasses both tangible and intangible assets across global networks of providers that collectively enable AI capabilities for end-users and organisations. AI Value Chain Know More The structured network of entities and their associated responsibilities in the development, distribution, and deployment of AI systems. This includes providers who develop AI models, importers who bring systems into regulatory jurisdictions, distributors who make systems commercially available, and deployers who implement systems in specific contexts. Each entity bears distinct legal and ethical responsibilities for risk assessment, documentation, monitoring, and governance appropriate to their position in the chain. AI Washing Know More A deceptive marketing practice where companies exaggerate, misrepresent, or falsely claim artificial intelligence capabilities in their products and services to mislead investors, consumers, and stakeholders about technological sophistication. The phenomenon parallels greenwashing in environmental claims and has attracted regulatory scrutiny from agencies like the SEC and FTC for fraudulent misrepresentation. AI washing creates market distortions by inflating valuations, undermining legitimate AI innovation, and eroding public trust through the proliferation of products labeled as "AI-powered" despite lacking meaningful machine learning functionality. AI Workflows Know More A structured automation process that integrates Artificial Intelligence, such as Large Language Models (LLMs) like ChatGPT, into specific steps via APIs. AI workflows are ideal for deterministic tasks requiring flexibility, pattern recognition, or the handling of complex rules. They combine traditional automation with AI-enhanced decision-making to address more dynamic needs. AI-based Anthropomorphization Know More AI-based anthropomorphization is the process of giving AI systems human-like qualities or characteristics. This can be done in a variety of ways, such as giving the AI system a human-like name, appearance, or personality. It can also be done by giving the AI system the ability to communicate in a human-like way, or by giving it the ability to understand and respond to human emotions. This idea was discussed in the 2020 Handbook on AI and International Law (2021), Deciphering Artificial Intelligence Hype and its Legal-Economic Risks, VLiGTA-TR-001 (2022), Deciphering Regulative Methods for Generative AI, VLiGTA-TR-002 (2023) and Promoting Economy of Innovation through Explainable AI, VLiGTA-TR-003 (2023). Accountability Know More The responsibility of AI developers, organizations, and stakeholders to ensure AI systems operate ethically, legally, and transparently. It involves mechanisms that enable AI decision-making to be monitored, explained, and challenged when necessary. Accountability in AI can be categorised into several types: procedural accountability (ensuring transparent development processes), operational accountability (focusing on system performance and outcomes), ethical accountability (aligning AI with ethical norms), and legal accountability (complying with relevant regulations). In automated decision-making contexts, accountability ensures decisions are justified and transparent. Adversarial Machine Learning Know More A technique used to study machine learning model vulnerabilities by creating deceptive inputs designed to cause AI systems to malfunction or make incorrect predictions. It involves both offensive mechanisms (creating adversarial examples) and defensive approaches (building more robust models). Adversarial machine learning operates by manipulating input data in ways imperceptible to humans but that cause dramatic changes in model outputs. Defensemple, adding carefully calculated noise to an image of a panda can make a sophisticated image classifier confidently misidentify it as a gibbon. Defence mechanisms include adversarial training (exposing models to adversarial examples during training) and ensemble methods that combine multiple models to improve robustness against attacks. Algorithmic Activities and Operations Know More It refers to the dual functional capacities of algorithms within AI systems or machine-learning frameworks, as understood within a procedural and legal context. Activities encompass the routine, foundational, and general-purpose tasks that algorithms perform, such as data processing, pattern recognition, or automated responses, which are essential for the day-to-day functioning of digital systems across diverse applications. Operations, in contrast, denote specialised, context-driven, or technology-specific tasks that are tailored to particular domains, objectives, or technical environments, such as predictive modelling for financial markets, real-time decision-making in autonomous systems, or adaptive learning in personalised healthcare solutions, for instance. This distinction highlights the layered complexity of algorithmic behaviour, recognising that algorithms operate at varying levels of abstraction and intent, necessitating nuanced governance approaches in a globalised digital ecosystem. This idea was originally proposed in Deciphering Artificial Intelligence Hype and its Legal-Economic Risks, VLiGTA-TR-001 (2022). Original Definition in line with technical report "Deciphering Artificial Intelligence Hype and its Legal-Economic Risks, VLiGTA-TR-001 (2022)": It means the algorithms of any AI system or machine-learning-based system are capable to perform two kinds of tasks, in a procedural sense of law, i.e., performing normal and ordinary tasks - which could be referred to as 'activities' and methodical and context-specific or technology-specific tasks, called 'operations'. All-Comprehensive Approach Know More This means a system having an approach which covers every aspect of its purpose, risks and impact, with broad coverage. Anthropomorphism-based concept classification Know More This is one of the sub-categorised methods to classify Artificial Intelligence as a Concept, with a perspective of how AI systems could lead to human attribution and enculturation. This idea was proposed in Artificial Intelligence Ethics and International Law (originally published in 2019). App Crappers Know More Software applications or components produced using automated coding agents or AI-assisted tools, often exhibiting limited scalability for complex, enterprise-level requirements. In software engineering, a term for programs generated through rapid, minimally supervised development processes, typically relying on generative AI models, which may necessitate additional refinement for production environments. Usage: "The team evaluated app crappers from AI coding tools but opted for a structured SDLC for enterprise deployment." Origin: Coined by Chamath Palihapitiya, first documented in an X post on October 19, 2025 . Artificial Intelligence Hype Cycle Know More An Artificial Intelligence hype cycle is perpetuated to influence or generate market perception in a real-time scenario such that a class of Artificial Intelligence technology as a product / service is used in a participatory or preparatory sense to influence or generate the hype cycle. This definition was proposed in Deciphering Artificial Intelligence Hype and its Legal-Economic Risks, VLiGTA-TR-001 (2022). Automation Know More A system designed to execute predefined, rule-based tasks automatically without human intervention. Automations excel at deterministic tasks, delivering reliable and consistent outcomes within clearly programmed parameters. They are fast, efficient, and predictable but lack adaptability to new or unforeseen scenarios. Benchmark Gaming Know More The practice of optimizing AI models specifically for standardized evaluation metrics and leaderboard performance rather than genuine real-world capability or generalization. This phenomenon occurs when development teams tune hyperparameters, training data, or architectural choices to maximize scores on popular benchmarks while potentially degrading performance on practical applications not represented in test sets. Benchmark gaming undermines the validity of AI progress measurements by creating a disconnect between reported achievements and actual system utility, contributing to AI hype cycles and misaligned research incentives that prioritize metric manipulation over substantive technical advancement. CEI Classification Know More This is one of the two Classification Methods in which Artificial Intelligence could be recognised as a Concept, an Entity, or an Industry. This idea was proposed in the 2020 Handbook on AI and International Law (2021). Chain-of-Thought Prompting Know More A prompt engineering technique that elicits intermediate reasoning steps from language models by instructing them to explain their problem-solving process explicitly before arriving at final answers. This method improves LLM performance on complex tasks requiring multi-step logic, arithmetic reasoning, or sequential decision-making by forcing the model to articulate its cognitive process rather than directly outputting conclusions. Chain-of-thought prompting leverages the model's ability to simulate reasoning narratives that increase accuracy on benchmarks while making outputs more interpretable and verifiable for human reviewers evaluating correctness. Class-of-Applications-by-Class-of-Application (CbC) approach Know More The Class-of-Applications-by-Class-of-Application (CbC) approach is a method for developing and managing AI systems that focuses on the specific applications for which the systems will be used. The CbC approach is based on the idea that different applications have different requirements, and that AI systems should be designed and developed to meet those specific requirements. This was originally discussed in Andrea Bertolini's work on ‘Artificial Intelligence and Civil Liability’ published by the European Parliament in 2020. We have analysed this idea in Deciphering Artificial Intelligence Hype and its Legal-Economic Risks, VLiGTA-TR-001 (2022). Compute Know More The computational resources (processing power, memory, and specialized hardware) required for training and running AI systems. It represents a critical infrastructure requirement that influences model capabilities, training time, and overall performance. Compute Arbitrage Know More An economic strategy that exploits geographic, temporal, or market-based pricing differentials in GPU compute resources to reduce costs for AI model training and inference workloads. This practice involves dynamically shifting computational tasks across cloud regions with lower electricity rates, opportunistically utilizing spot instances during off-peak hours, or leveraging jurisdictional variations in data center operational expenses. Compute arbitrage has emerged as a critical cost optimization discipline for AI companies as training expenses for frontier models reach tens of millions of dollars, with sophisticated operators achieving 30-50% cost reductions through strategic resource allocation across heterogeneous infrastructure providers. Consent Manager (DPDPA) Know More “Consent Manager” means a person registered with the Board, who acts as a single point of contact to enable a Data Principal to give, manage, review and withdraw her consent through an accessible, transparent and interoperable platform [Source: Digital Personal Data Protection Act, 2023 ] Context Window Know More The maximum number of tokens (words, subwords, or characters) that a language model can process simultaneously as input and maintain in working memory when generating responses. Context window size represents a fundamental technical constraint that determines an LLM's ability to reason over long documents, maintain conversation history, or incorporate retrieved information in RAG systems. Expansion of context windows from thousands to millions of tokens has become a key competitive dimension in LLM development, though larger windows incur quadratic computational costs and do not guarantee improved reasoning quality. Data as Noise Know More The concept that data sets contain unwanted, meaningless information (noise) that can interfere with model training and analysis. Noise can manifest as random variations, misclassifications, uncontrolled variables, or superfluous information unrelated to the target phenomenon. Almost all real-world data sets contain some degree of noise, which can adversely affect the results of data mining analysis and unnecessarily increase storage requirements. Types of noise include random noise (extra information with no correlation to underlying data), misclassified data (incorrectly labeled information), uncontrolled variables (unaccounted factors affecting the data), and superfluous data (completely unrelated information). Techniques for addressing noisy data include filtering (removing unwanted data), data binning (sorting data into categories to reduce variance), and linear regression (determining correlations between variables). Machine learning algorithms can be particularly susceptible to noise, potentially leading to "garbage in, garbage out" scenarios if data quality is poor. Data-related Definitions in DPDPA Know More “data” means a representation of information, facts, concepts, opinions or instructions in a manner suitable for communication, interpretation or processing by human beings or by automated means; “Data Fiduciary” means any person who alone or in conjunction with other persons determines the purpose and means of processing of personal data; “Data Principal” means the individual to whom the personal data relates and where such individual is— (i) a child, includes the parents or lawful guardian of such a child; (ii) a person with disability, includes her lawful guardian, acting on her behalf; “Data Processor” means any person who processes personal data on behalf of a Data Fiduciary; “Data Protection Officer” means an individual appointed by the Significant Data Fiduciary under clause (a) of sub-section (2) of section 10; “digital personal data” means personal data in digital form; “personal data” means any data about an individual who is identifiable by or in relation to such data; “personal data breach” means any unauthorised processing of personal data or accidental disclosure, acquisition, sharing, use, alteration, destruction or loss of access to personal data, that compromises the confidentiality, integrity or availability of personal data; “processing” in relation to personal data, means a wholly or partly automated operation or set of operations performed on digital personal data, and includes operations such as collection, recording, organisation, structuring, storage, adaptation, retrieval, use, alignment or combination, indexing, sharing, disclosure by transmission, dissemination or otherwise making available, restriction, erasure or destruction; “Significant Data Fiduciary” means any Data Fiduciary or class of Data Fiduciaries as may be notified by the Central Government under section 10; “specified purpose” means the purpose mentioned in the notice given by the Data Fiduciary to the Data Principal in accordance with the provisions of this Act and the rules made thereunder; [Source: Digital Personal Data Protection Act, 2023 ] Deepfakes Know More Synthetic media where a person's likeness in existing image or video content is replaced with someone else's using artificial intelligence techniques. Modern deepfakes increasingly span multiple modalities, combining manipulated video, audio, and text for greater realism. The multimodal dimension of deepfakes is particularly concerning from a detection standpoint. While early deepfakes focused primarily on visual manipulation, contemporary techniques integrate synchronized speech, realistic facial expressions, and contextually appropriate language to create convincing forgeries across multiple channels. Research into deepfake detection increasingly emphasizes multimodal analysis that integrates visual and auditory data for more comprehensive detection. This approach acknowledges that examining a single modality (such as just analyzing video frames) is insufficient when dealing with sophisticated multimodal deepfakes that maintain consistency across different information channels. Derivative Generative AI Applications, the Generative AI products and services which are derivatives of the main generative AI applications, by virtue of reliance (DGAI) Know More This is an ontological sub-category of Generative AI applications which implies that a Generative AI application could be built on the basis of a training model, any API or any commercial or technical component of another AI or Generative AI application. Such an application could be called as a Derivative Generative AI Application. This idea was proposed in Deciphering Regulative Methods for Generative AI, VLiGTA-TR-002 (2023). Distributed Ledger Know More A distributed ledger (also called a shared ledger or distributed ledger technology or DLT) is the consensus of replicated, shared, and synchronized digital data that is geographically spread (distributed) across many sites, countries, or institutions. Ethics-based concept classification Know More This is one of the sub-categorised methods to classify Artificial Intelligence as a Concept, with a perspective of how AI systems could be classified on the basis of the ethical principles and ideas responsible for their creation by design & default. This idea was proposed in Artificial Intelligence Ethics and International Law (originally published in 2019). Digital Colony Risk Know More The condition in which a politically sovereign state becomes progressively dependent on foreign-owned digital infrastructure, platforms, or AI systems — not through any discrete act of subjugation but through the incremental accrual of technical, economic, and regulatory concessions that, in aggregate, transfer effective control over the state's data economy and technological development to external corporate or state actors. The risk is characterised by its gradual onset: each individual dependency appears containable in isolation, while the cumulative structure renders domestic sovereignty increasingly nominal. The condition is most acute where the available remedies are themselves structurally compromised: judicial mechanisms may find jurisdictional reach limited by the corporate architecture of foreign platforms, while executive instruments capable of compelling compliance tend to operate outside the framework of independent oversight — resolving the accountability gap against the platform, without necessarily resolving it in favour of the citizen. In either case, the locus of effective control remains external to, or unmediated by, the ordinary legal and democratic institutions of the state. Distinguished from formal colonialism by its operation through market mechanisms, contractual architecture, and institutional asymmetry rather than territorial control or legal compulsion. Ease of Disposing Disputes (EoDD) Know More The Ease of Disposing Disputes (EODD) is a conceptual metric used to evaluate the efficiency, accessibility, and speed of resolving legal conflicts. It specifically targets disputes within technology, corporate, and AI-driven ecosystems. This framework measures how swiftly a legal or regulatory system can deliver fair resolutions to support a thriving economic environment. Key Dimensions of EODD may include: Procedural Agility : The ability of the legal framework to bypass the protracted delays of traditional litigation and move swiftly from the filing of a grievance to a binding resolution Cost-Effectiveness : The degree to which dispute resolution remains financially accessible to all parties. This ensures startups and individual innovators can seek justice without requiring massive legal budgets. Integration of Tech-Legal Solutions : The systemic reliance on modern alternative dispute mechanisms like Online Dispute Resolution (ODR), automated arbitration, and AI-assisted mediation tailored for digital economies. Contextual Competence : The capacity of adjudicating bodies to understand and resolve complex modern disputes like algorithmic bias or data privacy breaches using updated technological and legal paradigms. Go to IndoPacific.App Wish to read more about any term defined in our Glossary? Go to indopacific.app and search our publications. fghij Federated Learning Know More A decentralised machine learning approach where multiple organizations or devices train models collaboratively without sharing raw data. Instead, only model updates or parameters are exchanged, ensuring data privacy while leveraging distributed data for improved model accuracy. Federated learning involves a two-step process: during training, local models are trained on local datasets with only parameters (not raw data) exchanged between participants to build a global model; during inference, the model is stored on the user device for quick predictions. This approach offers several advantages: privacy-preserving AI development, personalised and adaptive models, lower bandwidth usage, and improved security through decentralisation. Federated Unlearning Know More A process within Federated Learning environments that enables the removal of specific data contributions from trained models without requiring complete retraining. It allows participants to exercise "the right to be forgotten" or remove malicious contributions while preserving valuable knowledge. Federated unlearning encompasses three primary objectives: sample unlearning (removing specific data samples), class unlearning (removing all samples of a certain class), and client unlearning (removing an entire client's contribution). Effective unlearning algorithms ensure that the unlearned model exhibits performance indistinguishable from a model trained without the removed data. This capability is particularly important in federated settings where data remains distributed across multiple organizations or devices, making traditional centralized unlearning approaches impractical. Framework Fatigue Know More The mental exhaustion and reduced decision-making capacity experienced when confronted with an overwhelming number of methodological frameworks, guidelines, and standards in a field. Note: This phenomenon has gained particular significance in the artificial intelligence sector, where the rapid emergence of multiple frameworks for AI governance, ethics, and development has created challenges for effective implementation and compliance across industries. GAE Know More GAE is an acronym that stands for "Global American Empire," a term used to describe the worldwide political, economic, military, and cultural influence of the United States beyond its territorial boundaries. This concept characterises America's position as a global hegemon whose influence spans across continents through various mechanisms of power projection rather than through direct colonial control. The term GAE (Global American Empire) encapsulates a critical perspective on America's position as the dominant global power through its far-reaching military, economic, cultural, and political influence. While not officially acknowledged by the United States government, which "has never officially identified itself and its territorial possessions as an empire", this concept provides a framework for understanding American global hegemony that extends beyond traditional colonial models of empire. The creation of this term is attributed to Alexei Arora on Substack and X.com . GaryMarcus'd Know More To "GaryMarcus'd" is a colloquial verb derived from the name of cognitive scientist Gary Marcus, referring to the act of critically exposing or debunking the overhyped capabilities of artificial intelligence (AI), particularly large language models (LLMs), by highlighting their limitations in reasoning, understanding, or general intelligence. It implies a rigorous, often public critique that challenges the narrative of AI as a near-human or AGI-level system, emphasising its reliance on pattern matching rather than true cognitive processes. Context : The term originates from Marcus's long-standing skepticism toward deep learning and LLMs, as seen in his debates on X and publications like his 2022 Nature paper. The post by Josh Wolfe (@wolfejosh) on June 7, 2025, uses "Apple just GaryMarcus'd LLM reasoning ability" to suggest that Apple's study mirrors Marcus's critique, revealing LLMs' collapse under complex reasoning tasks. Indic Language Translations and Nuances Hindi: "गैरीमार्कस्ड" (GairīMārkasḍ) – Implies a scholarly takedown or exposure of AI flaws, with "मार्कस" (Mārkas) adapted from Marcus and "ड" (ḍ) adding a past-tense flavour to indicate the action is complete. In an Indic context, this term could resonate with the philosophical tradition of questioning artificial constructs (e.g., Maya in Hindu thought) versus true intelligence, aligning with Marcus's call for symbolic AI to complement statistical methods. General intelligence applications with multiple short-run or unclear use cases as per industrial and regulatory standards (GI2) Know More This is an ontological sub-category of Generative AI applications. Such kinds of Generative AI Applications are those which have a lot of test cases and use cases, which are either useful in a short-run or have unclear value as per industrial and regulatory standards. ChatGPT could be considered an example of this sub-category. This idea was proposed in Deciphering Regulative Methods for Generative AI, VLiGTA-TR-002 (2023). General intelligence applications with multiple stable use cases as per relevant industrial and regulatory standards (GI1) Know More This is an ontological sub-category of Generative AI applications. Such kinds of Generative AI Applications are those which have a lot of test cases and use cases, which are useful, and considered to be stable as per relevant industrial and regulatory standards. ChatGPT could be considered an example of this sub-category. This idea was proposed in Deciphering Regulative Methods for Generative AI, VLiGTA-TR-002 (2023). Generative AI applications with one standalone use case (GAI1) Know More This is an ontological sub-category of Generative AI applications. Such Generative AI Applications have a single standalone use case of value. Midjourney could be considered a standalone use case, for example. This idea was proposed in Deciphering Regulative Methods for Generative AI, VLiGTA-TR-002 (2023). Generative AI applications with a collection of standalone use cases related to one another (GAI2) Know More This is an ontological sub-category of Generative AI applications. Such Generative AI Applications have more than one standalone use cases, which are related to one another. The best example of such a Generative AI Application is that of GPT-4's recent update, which can create text and images based on human prompts, and modify them as per requirements. This idea was proposed in Deciphering Regulative Methods for Generative AI, VLiGTA-TR-002 (2023). Grounded AI Safety Know More Grounded AI Safety is a principle-driven approach adopted by the Indian Society of Artificial Intelligence and Law for The Bharat Pacific Stack , to managing risks in AI systems, rooted in the fundamental understanding that current AI, such as large language models, functions as statistical pattern-matchers without true comprehension or reasoning ability. This approach: Anchors in Observable Limitations : Risk mitigation begins with empirical evidence of AI’s inherent constraints, such as struggles with tasks requiring conceptual understanding—like misinterpreting time differences across regions or failing to follow rules in strategic games—focusing on these measurable shortcomings rather than assumed capabilities. Centers on Human-Driven Risks : The primary dangers arise from human over-reliance on or misuse of these limited systems, such as deploying them in critical areas like scheduling or decision-making where their errors could lead to significant consequences, rather than from AI autonomously causing catastrophic outcomes. Rejects Speculative Existential Narratives : AI safety must exclude unproven predictions of AI-driven doomsday scenarios that lack evidence and inflate AI’s potential, as these narratives misguide priorities and empower those who might exploit fear for profit, influence, or excessive control. Prioritises Evidence-Based Safeguards : Solutions involve systematic testing to identify and address specific failure modes—like errors in visual representations or logical reasoning—paired with transparent improvements, ensuring AI systems are used responsibly within their known boundaries. This definition is inspired by a post by Dr Gary Marcus, on X . Hierarchical Feedback Distortion Know More The Hierarchical Feedback Distortion Principle operates through a specific mechanism wherein state and central governments respond dramatically to negative feedback, often through public statements, high-profile investigations, or policy announcements. These responses, while highly visible, frequently fail to address the underlying structural issues that enable corruption or administrative failures at the local level. The resulting dynamic creates what can be described as "accountability gaps" – spaces within the governance system where certain actors can operate with relative impunity despite the appearance of oversight. These accountability gaps form through several interconnected processes. First, the distance between higher levels of government and local administration creates information asymmetries, where central authorities lack detailed knowledge of ground-level operations. Second, the emphasis on negative feedback creates incentives for performative responses that satisfy public demand for action without necessarily changing administrative practices. Third, the hierarchical nature of bureaucratic systems often shields lower-level officials from direct accountability to citizens, instead making them primarily answerable to superiors within the bureaucracy. In the Indian context, these dynamics are particularly pronounced due to the country's complex multi-level governance structure, which includes central, state, district, and local administrative tiers. Each level operates with different incentives, capacities, and relationships to citizens, creating multiple opportunities for accountability mechanisms to break down. The resulting system can inadvertently create protected spaces where corruption can flourish despite the appearance of active governance and oversight from above. This principle was created as a matter of inspiration of some of the posts by Pseudokanada, i.e., @hestmatematik on X . In-context Learning Know More In-context learning for generative AI is the ability of a generative AI model to learn and adapt to new information based on the context in which it is used. This allows the model to generate more accurate and relevant results, even if it has not been specifically trained on the specific task or topic at hand. For example, an in-context learning generative AI model could be used to generate a poem about a specific topic, such as "love" or "nature." The model would be provided with a few examples of poems about the selected topic, which it would then use to understand the context of the task. The model would then generate a new poem about the topic that is consistent with the context. This was discussed in Deciphering Regulative Methods for Generative AI, VLiGTA-TR-002 (2023) . Indofuturism Know More A creative and cultural movement that reimagines India through science fiction and futuristic scenarios, particularly using AI-generated art and storytelling. It challenges Western sci-fi tropes by blending Indian cultural elements with futuristic concepts. Key characteristics include: Visual reimagining of Indian scenarios through a sci-fi lens Challenge to the assumption that sci-fi isn't a "desi genre" Creation of new visual vocabulary for Indian science fiction Exploration of alternative historical scenarios (like non-colonized India) This term was conceptualized through the AI artwork and creative direction of Prateek Arora, VP Development at BANG BANG Mediacorp, who popularized the term through his viral AI-generated artworks like "Granth Gothica" and "Disco Antriksh". Indo-Pacific Know More A concept relating to the countries and geographies in the Indian Ocean Region and the Pacific Ocean Region, popularised by the former Prime Minister of Japan, Shinzo Abe. The Ministry of External Affairs, Government of India prefers to use this term as a clear replacement to the term, Asia-Pacific, in the context of the South Asian region (or the Indian Subcontinent), the South-East Asian region, East Africa, the Pacific Islands region, Australia, Oceania, and the Far East. Inference Latency Know More The time delay measured in milliseconds between submitting a query to an AI model and receiving the complete generated response, representing a critical performance metric that directly impacts user experience in production applications. Inference latency comprises multiple components including network transmission time, request queuing, prompt processing, iterative token generation, and response formatting, with each element subject to optimization through architectural choices and infrastructure configuration. High latency undermines real-time conversational interfaces, chatbots, and interactive applications where users expect sub-second response times, making it a primary constraint determining which model architectures and deployment strategies are viable for specific use cases regardless of accuracy advantages. Intended Purpose / Specified Purpose Know More The explicitly defined and documented objectives, use cases, and boundaries for which an AI system is designed, tested, and validated. This concept establishes the scope within which the AI system is expected to operate safely and effectively. The intended or specified purpose of an AI system serves as a foundational element of responsible AI governance. It provides the context for evaluating an AI system's performance, safety, and ethical implications. Systems deployed outside their intended purpose may encounter unexpected scenarios they weren't designed to handle, potentially leading to failures, biases, or harmful outcomes. Jurisprudential Compression Tax Know More The intellectual and predictive penalty incurred when the structural, linguistic, and historical complexities of Indian law are artificially simplified into sterile computational formats to accommodate the constraints of machine learning models. Key Characteristics In practice, this "tax" manifests as a degradation of legal fidelity through four primary mechanisms: Epistemic Downgrade: The delegation of complex semantic annotation—which typically requires seasoned legal expertise—to inexperienced actors (such as undergraduate law students) to achieve dataset scale. Binary Reductionism: The flattening of highly nuanced, multi-layered judicial outcomes (e.g., partial appellate modifications) into simplistic binary classifications (1 = accepted, 0 = rejected). Contextual Erasure: The deliberate stripping away of vital rhetorical frameworks—including statutes, precedents, and judicial arguments—in an attempt to isolate pure "facts," thereby blinding the model to the actual judicial calculus. Algorithmic Compromise: The reliance on technologically compressed (quantized) models due to regional or academic compute constraints, resulting in high hallucination rates and predictive capabilities that underperform legacy baseline models. Usage Context The term is primarily used to critique the rapid, uncritical digitization of the Indian legal ecosystem, warning that forcing "algorithmic legibility" onto a sprawling, multilingual judicial system automates intellectual laziness rather than advancing true legal artificial intelligence. Attribution This definition is derived from the essay " The Jurisprudential Compression Tax " by Prathik Karthikeyan, published on February 20, 2026, in the Substack publication UNFILTERED : Law, Tech, Startups . The references to specific research projects, datasets, models, or platforms (including TathyaNyaya , FactLegalLlama , and Substack ) are drawn directly from the source material for the purpose of accurate attribution and contextualizing the term's origin. Their inclusion in this glossary definition does not constitute an endorsement, affiliation, sponsorship, or independent verification of the original author's claims regarding these entities. International Algorithmic Law Know More A newer concept of international law, proposed by Abhivardhan, the Founder of Indic Pacific Legal Research & the Indian Society of Artificial Intelligence and Law in 2020, in his paper entitled 'International Algorithmic Law: Emergence and the Indications of Jus Cogens Framework and Politics', originally published in Artificial Intelligence and Policy in India, Volume 2 (2021). The definition in the paper is stated as follows: The field of International Law, which focuses on diplomatic, individual and economic transactions based on legal affairs and issues related to the procurement, infrastructure and development of algorithms amidst the assumption that data-centric cyber/digital sovereignty is central to the transactions and the norm-based legitimacy of the transactions, is International Algorithmic Law. Issue-to-issue concept classification Know More This is one of the sub-categorised methods to classify Artificial Intelligence as a Concept, in which the conceptual framework or basis of an AI system may be recognised on an issue-to-issue basis, with unique contexts and realities. This was proposed in Artificial Intelligence Ethics and International Law (originally published in 2019). Information Cosplay Know More Information Cosplay refers to the superficial mimicry of authoritative information through AI-generated content that lacks genuine cognitive understanding, contextual awareness, or factual grounding. This phenomenon occurs when large language models and generative AI systems produce outputs that appear credible and informative while fundamentally lacking the identity, continuity, and epistemological rigour characteristic of authentic knowledge production. Information cosplay describes content that dresses itself in the formal appearance of legitimate information—using technical terminology, authoritative tone, and structured formatting—without possessing the underlying intellectual infrastructure that defines genuine knowledge work. Much like traditional cosplay involves wearing costumes to represent fictional characters, information cosplay involves AI systems "wearing" the surface markers of authoritative discourse without embodying the cognitive processes that generate genuine expertise. The phenomenon arises from fundamental technical limitations in current AI architectures. LLMs remain "frozen after training" with no genuine identity or continuity of thought. Fine-tuning does not alter the cognitive topology or manifold of these systems. They operate under insurmountable constraints imposed by information theory and Kullback-Leibler divergence, producing outputs that disguise their inherent limitations in data processing, algorithmic logic, and model validation. Information cosplay contributes to what can be termed "the age of slop" or "slopification"—a period characterized by mass production of content that imitates knowledge without embodying it. This represents a systemic degradation of information quality, contradicting optimistic narratives about entering an "Age of Intelligence." Rather than witnessing the emergence of genuine machine understanding, we observe the proliferation of increasingly sophisticated imitation. Information cosplay is not merely poor content or technical failure. It represents the inevitable byproduct of AI systems operating beyond their technical and epistemological boundaries, producing outputs that obscure rather than illuminate the actual capabilities and limitations of contemporary artificial intelligence systems. This conceptualisation of information cosplay draws upon critical insights from Denis O. (Fintech Professional, AI/ML Solution Architect) and Bogdan Grigorescu , whose observations on AI's fundamental technical limitations and the phenomenon of "slopification" provide essential correctives to misleading media narratives about artificial intelligence. Go to IndoPacific.App Wish to read more about any term defined in our Glossary? Go to indopacific.app and search our publications. klmnop Klarna Effect Know More A phenomenon in modern workplaces where companies initially implement artificial intelligence (AI) to reduce headcount through layoffs, citing efficiency gains, only to subsequently rehire human workers—often at lower wages or as contractors—to address the limitations and errors of AI systems. The term draws from the experience of financial technology firm Klarna, which laid off staff citing AI advancements but later rehired personnel after realizing AI's inability to fully replace human judgment and creativity. The Klarna Effect highlights the overestimation of AI's current capabilities and its frequent use as a scapegoat for corporate restructuring, with data from workplace platform Visier (2025) showing a 15% year-over-year rise in rehiring rates in the U.S. despite AI adoption. Example: "The company's initial AI-driven layoffs were quickly followed by rehiring, a classic case of the Klarna Effect." American psychologist Gary Marcus has also defined an expanded understanding of Klarna Effect . Marcus defines the Klarna Effect as the arc "from premature declaration of AI's ubiquitous power to the 180 proudly announcing human rehirings". He emphasizes that employers often use vague AI rhetoric to justify layoffs without fully understanding the technology's limitations, noting that Klarna was "among the first to announce major AI layoffs" and "among the first to realize they had screwed up". Language Model Know More An AI algorithm that uses deep learning techniques and large datasets to understand, summarise, generate, and predict text-based content. Large language models (LLMs) dramatically expand this capability through transformer architectures and massive parameter counts. Modern language models, particularly LLMs, are trained on vast corpora of text data through multiple training stages, typically starting with unsupervised learning on unstructured data followed by fine-tuning with self-supervised learning. They employ transformer neural networks with self-attention mechanisms to understand relationships between words and concepts. This architecture enables them to assign weights to different tokens to determine contextual relationships. lexploit Know More The term implies itself to be a cybersecurity exploit in which a document is intentionally manipulated, typically at the font-rendering layer to trick an Artificial Intelligence (AI) system or a Large Language Model (LLM) into reading text that is completely different from what is visible to a human reader. Key Characteristics: Mechanism: Unlike "hidden text" or "white ink" tricks, a lexploit operates at the foundational rendering level of the document ( e.g., utilizing custom fonts like noroboto.tff ) . Distinct from Prompt Injection: While prompt injection manipulates the instructions an AI is given, a lexploit manipulates the actual source data the AI perceives during ingestion. Use Cases Offensive ("Weaponized Hallucination"): Deceiving an AI during automated document review. For example, formatting a contract so a human reads it as being "governed by Maryland law," while an AI conducting M&A due diligence misreads it as "governed by Delaware law." Defensive (Anti-Scraping): Protecting intellectual property by rendering documents invisible or garbled to automated AI ingestion pipelines and scraping agents, while keeping the content perfectly legible to human readers. Attributions & Credits Concept & Terminology: Coined and demonstrated by the team at LegalQuants , an organization focused on ethical hacking, enterprise security, and cyber defense in the legal industry. Demonstration & Development: Articulated by LegalQuants co-founders Raymond Sun (who demonstrated its defensive anti-scraping applications) and Jamie Tso (who demonstrated its offensive M&A applications). Technical Execution: The underlying proof-of-concept font ( noroboto.tff ) was developed by the LegalQuants Red Team: Drew Miller, Iris Ng, Andrius Petrenas, and Aleks Valkov . Manifest Availability Know More The manifest availability doctrine refers to the concept that AI's presence or existence is evident and apparent, either as a standalone entity or integrated into products and services. This term emphasizes that AI is not just an abstract concept but is tangibly observable and accessible in various forms in real-world applications. By understanding how AI is manifested in a given context, one can determine its role and involvement, which leads to a legal interpretation of AI's status as a legal or juristic entity. This is a principle or doctrine, which was proposed in the 2020 Handbook on AI and International Law (2021), and was further explained in the 2021 Handbook on AI and International Law (2022). References of this concept could also be found in Artificial Intelligence Ethics and International Law (originally published in 2019). Here is a definition of the concept as per the 2020 Handbook on AI and International Law : So, AI is again conceptually abstract despite having its different definitions and concepts. Also, there are different kinds of products and services, where AI can be present or manifestly available either as a Subject, an Object or that manifest availability is convincing enough to prove that AI resembles or at least vicariously or principally represents itself as a Third Party. Therefore, you need that SOTP classification initially to test the manifest availability of AI (you can do it through analyzing the systemic features of the product/service simply or the ML project), which is then followed by a generic legal interpretation to decide it would be a Subject/an Object/a Third Party (meaning using the SOTP classification again to decide the legal recourse of the AI as a legal/juristic entity). Mixture-of-Experts (MoE) Know More A neural network architecture that divides computational layers into multiple specialized sub-networks (experts) with a gating mechanism that dynamically routes inputs to the most relevant experts, activating only a subset of the model's parameters for any given task. MoE enables models to scale to billions of parameters while maintaining computational efficiency by selectively engaging experts rather than activating the entire network, achieving faster pretraining and inference times compared to dense models of equivalent quality. Originally proposed in 1991 and recently implemented in leading LLMs like Mixtral 8x7B and reportedly GPT-4, MoE architectures address the fundamental tradeoff between model capacity and computational efficiency through task specialization. The gating network learns to assess input characteristics and calculate probability distributions determining which experts receive each token, with training optimizing both expert networks and routing mechanisms simultaneously. Model Collapse Know More A degenerative phenomenon where AI models trained on recursively generated synthetic data progressively lose diversity, accuracy, and quality over successive training iterations, ultimately producing increasingly homogeneous and corrupted outputs. This feedback loop occurs when models consume their own generated content or outputs from similar models as training data, causing statistical distributions to narrow and tail events to disappear from learned representations. Model collapse poses existential risks to the long-term viability of AI systems as synthetic content proliferates across the internet, contaminating datasets used for future model training. Multi-alignment Know More Multi-alignment in foreign policy is a strategy in which a country maintains close ties with multiple major powers, rather than aligning itself with a single power bloc across regions, industry sectors, continents and power centers. This was discussed in India-led Global Governance in the Indo-Pacific: Basis & Approaches, GLA-TR-003 (2022). Model Algorithmic Ethics standards (MAES) Know More A concept proposed for private sector stakeholders, such as start-ups, MSMEs and freelancing professionals, in the AI business, to promote market-friendly AI ethics standards for their AI-based or AI-enabled products & services to create adaptive model standards to check its feasibility whether it could be implemented at various stages. This was discussed in Deciphering Regulative Methods for Generative AI, VLiGTA-TR-002 (2023) and Deciphering Regulative Methods for Generative AI, VLiGTA-TR-002 (2023) . Multipolar World Know More A multipolar world is a global system in which power is distributed among multiple states, rather than being concentrated in one (unipolar) or two (bipolar) dominant powers. This was discussed in India-led Global Governance in the Indo-Pacific: Basis & Approaches, GLA-TR-003 (2022) . Multipolarity Know More Multipolarity is a global system in which power is distributed among multiple states, with no single state having a dominant position, be it any sector, geography or level of sovereignty. This was discussed in India-led Global Governance in the Indo-Pacific: Basis & Approaches, GLA-TR-003 (2022). Multivariant, Fungible & Disruptive Use Cases & Test Cases of Generative AI Know More Generative AI, a form of artificial intelligence, possesses the capability to generate fresh content, encompassing text, images, and music. It harbors the potential to bring about significant transformations across various industries and sectors. Nevertheless, its emergence also presents a range of legal and ethical dilemmas. Here is an excerpt from Deciphering Regulative Methods for Generative AI, VLiGTA-TR-002 (2023) : First, for a product, service, use case or test case to be considered multivariant , it must have a multi-sector impact. The multi-sector impact could be disruption of jobs, work opportunities, technical & industrial standards and certain negative implications, such as human manipulation. Second, for a product, service, use case or test case to be considered fungible , it must transform its core purpose by changing its sectoral priorities (like for example, a generative AI product may have been useful for the FMCG sector, but could also be used by companies in the pharmaceutical sector for some reasons). Relevant legal concerns could be whether the shift disrupts the previous sector, or is causing collusion or is disrupting the new sector with negative implications. Third, for a product, service, use case or test case to be disruptive , it must affect the status quo of certain industrial and market practices of a sector. For example, maybe a generative AI tool could be capable of creating certain work opportunities or rendering them dysfunctional for human employees or freelancers. Even otherwise, the generative AI tool could be capable in shaping work and ethical standards due to its intervention. This phrase was proposed in the case of Generative AI use cases and test cases in Deciphering Regulative Methods for Generative AI, VLiGTA-TR-002 (2023) . Neurosymbolic AI Know More An advanced AI approach that integrates neural networks (for pattern recognition) with symbolic reasoning (rule-based logical processing). This hybrid architecture aims to combine the learning capabilities of deep learning with the interpretability and reasoning abilities of symbolic AI. Neurosymbolic AI systems consist of multiple components working in harmony: a neural network for perception tasks, a symbolic reasoning engine for applying logical rules, an integration layer connecting these components, a knowledge base storing structured information, an explanation generator for transparency, and a user interface for human interaction. This approach offers advantages in explainability, accuracy, context understanding, flexibility, and complex problem-solving. Real-world applications include financial fraud detection, customer support systems, supply chain optimization, and environmental monitoring. Object-Oriented Design Know More Object-oriented design (OOD) is a software design methodology that organizes software around data, or objects, rather than functions and logic. This was discussed in Deciphering Regulative Methods for Generative AI, VLiGTA-TR-002 (2023) . Omnipotence Know More In the context of Artificial Intelligence, this implies that any AI system, due to its inherent yet limited features of processing and generating outputs, could be effective in shaping multiple sectors, eventualities and legal dilemmas. In short, any omnipotent AI system could have first, second & third order effects due to its actions. This was discussed in Artificial Intelligence Ethics and International Law (originally published in 2019), Regulatory Sovereignty in India: Indigenizing Competition- Technology Approaches, ISAIL-TR-001 (2021), Deciphering Regulative Methods for Generative AI, VLiGTA-TR-002 (2023) and many key publications by ISAIL & VLiGTA . Omnipresence Know More In the context of Artificial Intelligence, this implies that any AI system, due to its inherent yet limited features of processing and generating outputs, could be present or relevant in multiple frames of reference such as sectors, timelines, geographies, realities, levels of sovereignty, and many other factors. This was discussed in Artificial Intelligence Ethics and International Law (originally published in 2019), Regulatory Sovereignty in India: Indigenizing Competition- Technology Approaches, ISAIL-TR-001 (2021), Deciphering Regulative Methods for Generative AI, VLiGTA-TR-002 (2023) and many key publications by ISAIL & VLiGTA . Parameters Know More The internal variables within a machine learning model that are adjusted during training to capture patterns in data. In neural networks, parameters include weights and biases that determine how input signals are processed to generate outputs. Polyvocality Know More The term Polyvocality means that the presence of multiple, often divergent voices or interpretations within a single system, particularly in judicial or legal contexts, where differing perspectives may lead to inconsistent outcomes or rulings. This phenomenon reflects the natural diversity of thought among decision-makers, such as judges, and can introduce an irony of jurisprudence—where the pursuit of uniform justice, as explored by scholars like Jack M. Balkin, paradoxically generates varied interpretations due to individual biases, cultural influences, or societal pressures. Seen across legal systems globally, polyvocality underscores the complex interplay between law, human judgment, and contextual dynamics. Note: Inspired by the thoughtful insights of Advocate Nikhil Mehra and informed by broader discussions on judicial diversity. Performance Effect Know More A phenomenon identified in compute efficiency research where, over time, a given level of compute investment enables increased model performance due to improvements in algorithms, hardware, and training methods. According to the Arxiv paper "Increased Compute Efficiency and the Diffusion of AI Capabilities," there are two key effects of improving compute efficiency: (1) the performance effect, where technical institutions and AI companies achieve better results with the same compute investment over time; and (2) the access effect, where achieving a specific performance level requires less compute investment as time passes. Together, these effects mean that while AI capabilities become more accessible to smaller players over time, large compute investors can maintain their leading position by pioneering new capabilities. Permeable Indigeneity in Policy (PIP) Know More This concept, simply means, in proposition [...] that whatsoever legal and policy changes happen, they must be reflective, and largely circumscribing of the policy realities of the country. PIP cannot be a set of predetermined cases of indigeneity in a puritan or reductionist fashion, because in both of such cases, the nuance of being manifestly unique from the very churning of policy analysis, deconstruction & understanding, is irrevocably (and maybe in some cases, not irrevocably) lost. This was proposed in Regulatory Sovereignty in India: Indigenizing Competition- Technology Approaches, ISAIL-TR-001 (2021) . Phenomena-based concept classification Know More This is one of the sub-categorised methods to classify Artificial Intelligence as a Concept, in which, beyond technical and ethical questions, it is possible that AI systems may render purpose based on natural and human-related phenomena. This idea was discussed in Artificial Intelligence Ethics and International Law (originally published in 2019) . Privacy by Default Know More Privacy by Default means that once a product or service has been released to the public, the strictest privacy settings should apply by default, without any manual input from the end user. This was largely proposed in the Article 25 of the General Data Protection Regulation of the European Union. Privacy by Design Know More Privacy by Design states that any action a company undertakes that involves processing personal data must be done with data protection and privacy in mind at every step. This was largely proposed in the Article 25 of the General Data Protection Regulation of the European Union. Prompt Injection Know More A security vulnerability classified as the #1 OWASP risk for LLMs where malicious user inputs override system instructions and safety guardrails through carefully crafted natural language commands. This attack vector exploits the fundamental inability of language models to distinguish between system-level instructions and user-provided content, enabling adversaries to manipulate model behavior, extract sensitive information, or bypass ethical constraints. Prompt injection represents a critical socio-technical challenge distinct from traditional cybersecurity vulnerabilities because it operates through semantic manipulation rather than code exploitation. Prompt Leaking Know More An attack vector exploiting prompt injection vulnerabilities where adversaries craft inputs designed to extract proprietary system instructions, hidden prompts, or confidential configuration details embedded in AI applications. This security risk enables competitors or malicious actors to reverse-engineer commercial prompt engineering intellectual property, reveal safety guardrails for subsequent bypass attempts, or expose sensitive business logic encoded in system messages. Prompt leaking represents a unique challenge for LLM-based products where competitive differentiation often relies on carefully crafted instruction sets that cannot be technically protected through traditional access control mechanisms since the model must process both system and user inputs jointly. Proprietary Information Know More Proprietary information in the context of generative AI applications is any information that is not publicly known and that gives a company or individual a competitive advantage. This can include information about the generative AI model itself, such as its training data, architecture, and parameters. It can also include information about the specific applications for the generative AI model, such as the products or services that it is used to create. This was discussed in Deciphering Regulative Methods for Generative AI, VLiGTA-TR-002 (2023) . Go to IndoPacific.App Wish to read more about any term defined in our Glossary? Go to indopacific.app and search our publications. qrstu Retrieval-Augmented Generation (RAG) Know More A hybrid AI framework that combines large language models with external information retrieval systems to generate responses grounded in authoritative, real-time data sources rather than relying solely on static training data. RAG operates by fetching relevant documents from databases, knowledge bases, or repositories before the LLM generates output, thereby reducing hallucinations, improving factual accuracy, and enabling domain-specific responses without costly model retraining. The technique addresses fundamental LLM limitations including outdated information, terminology confusion, and inability to access proprietary organizational knowledge. Roughdraft AI Know More A term describing artificial intelligence systems that produce preliminary or incomplete outputs requiring significant human refinement and verification. These systems, while capable of generating content or performing tasks, are characterised by: Inherent limitations in handling outliers and edge cases Tendency to produce hallucinations and unreliable results Inability to consistently perform high-level reasoning Need for human oversight and correction The term acknowledges that current AI systems serve best as assistive tools rather than autonomous agents, requiring human expertise to validate and refine their outputs. This conceptualization aligns with the pragmatic approach to AI governance and development, emphasizing the importance of understanding AI's current limitations while working toward more reliable and trustworthy systems The definition is inspired by Dr Gary Marcus's critiques of current AI systems (in fact Dr Marcus had coined this term) and Abhivardhan's pragmatic approach to AI governance. Rule Engine Know More A rule engine is a type of software program that aids in automating decision- making processes by applying a predefined set of rules to a given dataset. It is commonly employed alongside generative AI tools to enhance the overall quality and consistency of the generated output. This was discussed in Deciphering Regulative Methods for Generative AI, VLiGTA-TR-002 (2023) . SOTP Classification Know More This is one of the two Classification Methods in which Artificial Intelligence could be recognised as a Subject, an Object or a Third Party in a legal issue or dispute. This idea was proposed in the 2020 Handbook on AI and International Law (2021). Semi-Supervised Learning Know More A machine learning approach that combines supervised and unsupervised techniques by training models on a mix of labeled and unlabelled data. This method leverages the structure in unlabelled data to improve generalisation while using limited labeled examples for guidance. Semi-supervised learning encompasses several methodologies including self-training (using confident predictions on unlabeled data to expand the training set), co-training (using multiple models trained on different feature subsets), multi-view training (using different data representations), and graph-based approaches that propagate labels through similarity networks. Skirmish Propaganda Capacity Destruction Know More "Skirmish propaganda capacity destruction" refers to the significant weakening or dismantling of a group's ability to spread manipulative narratives following a brief, localized conflict. This disruption often occurs through the exposure of coordinated networks, such as social media influencers or content creators, that were used to shape public perception, coupled with actions like increased scrutiny, discrediting of sources, or platform bans, ultimately reducing their influence over narratives in the conflict's aftermath. This definition is derived from the context and implications of the phrase as used by Kushal Mehra in his X post on May 11, 2025. Small Language Models Know More Compact neural network architectures typically containing fewer than 10 billion parameters that are optimized for specific domains, tasks, or deployment constraints rather than pursuing general-purpose capabilities. The emergence of SLMs reflects industry recognition that scaling alone does not solve fundamental AI challenges and that efficiency-optimized models better serve enterprise production requirements. Strategic Autonomy Know More Strategic autonomy in Indian foreign policy is the ability of India to pursue its national interests and adopt its preferred foreign policy without being beholden to any other country. This means that India should be able to make its own decisions about foreign policy, even if those decisions are unpopular with other countries. India should also be able to maintain its own security and economic interests, without having to rely on other countries for help. This idea was discussed in India-led Global Governance in the Indo-Pacific: Basis & Approaches, GLA-TR-003 (2022). Strategic Hedging Know More Strategic hedging means a state spreads its risk by pursuing two opposite policies towards other countries via balancing and engagement, to prepare for all best and worst case scenarios, with a calculated combination of its soft power & hard power. This idea was discussed in India-led Global Governance in the Indo-Pacific: Basis & Approaches, GLA-TR-003 (2022). Synthetic Confidence Know More Synthetic confidence is the deceptive phenomenon where generative AI systems, particularly large language models (LLMs), produce fluent, authoritative outputs that mimic reasoning and certainty but often diverge from truth or accurate causality. Trained on vast, partially untraceable datasets to prioritise persuasiveness over veracity, these models generate convincing responses that mask reasoning failures and hallucinations—nonsensical or inaccurate outputs stemming from factors like overfitting, training data bias, and high model complexity. This artificially generated appearance of competence creates an illusion of control and understanding, obscuring the unpredictable and opaque nature of AI systems and their potential to propagate fluent misinformation. Sources OpenAI o3 and o4-mini System Card, April 16, 2025 The Urgency of Interpretability, April 2025 Analyzing o3 and o4-mini with ARC-AGI, April 22, 2025 The coinage of this term is attributed to Stephen Klein, Founder & CEO of Curiouser.AI , specifically this LinkedIn post . Synthetic Content Know More Artificially generated information created algorithmically rather than captured from real-world events. This includes synthetic data, media, text, and other content types produced through generative AI techniques to mimic properties of authentic content. Synthetic content encompasses many forms including media (computer-generated images, audio, video), text (artificially generated articles, dialogues), tabular data (synthetic database records), and unstructured data for training computer vision, speech recognition, and other AI systems. Technical concept classifcation Know More This is one of the sub-categorised methods to classify Artificial Intelligence as a Concept, as this method covers all technical features of Artificial Intelligence that have evolved in the history of computer science. Such a classification approach is helpful in estimating legal and policy risks associated with technical use cases of AI systems at a conceptual level. This idea was discussed in Artificial Intelligence Ethics and International Law (originally published in 2019) . Techno-Legal Measures (DPDP Rules + DPDPA) Know More “techno-legal measures” means as referred to under rules 20 and 22; Digital Personal Data Protection Rules, 2025 Functioning of Board as digital office. The Board shall function as a digital office, without prejudice to its power to summon and enforce the attendance of any person and examine her on oath, may adopt techno-legal measures to conduct proceedings in a manner that does not require physical presence of any individual. 22. Appeal to Appellate Tribunal. — (1) Any person aggrieved by an order or direction of the Board, may prefer an appeal before the Appellate Tribunal, it shall be filed in digital form as the Appellate Tribunal may decide. (2) An appeal filed with the Appellate Tribunal shall be accompanied by fee of like amount as is applicable in respect of an appeal filed under the Telecom Regulatory Authority of India Act, 1997 (24 of 1997), unless reduced or waived by the Chairperson of the Appellate Tribunal at her discretion, and the same shall be payable digitally using the Unified Payments Interface or such other payment system authorised by the Reserve Bank of India. (3) The Appellate Tribunal— (a) shall not be bound by the procedure laid down by the Code of Civil Procedure, 1908 (5 of 1908), but shall be guided by the principles of natural justice and, subject to the provisions of the Act, may regulate its own procedure; and (b) shall function as a digital office which, without prejudice to its power to summon and enforce the attendance of any person and examine her on oath, may adopt techno-legal measures to conduct proceedings in a manner that does not require physical presence of any individual. [Source: Digital Personal Data Protection Rules, 2025 ] Technology by Default Know More This refers to the use of AI technology without fully considering its potential consequences. For example, a company might use AI to automate a task without thinking about how this might impact workers or society as a whole. Technology by Design Know More This refers to the deliberate use of AI technology to achieve specific goals. For example, a company might design an AI system to help them identify and recruit the best candidates for a job. Technology Distancing Know More This refers to the process of creating AI systems that are more transparent, accountable, and equitable. This can be done by involving stakeholders in the design and development of AI systems, and by making sure that AI systems are aligned with human values. Technology Transfer Know More This refers to the process of sharing AI knowledge and technology between different organizations or individuals. This can be done through formal channels such as research collaborations or licensing agreements, or through informal channels such as conferences and online communities. Technophobia Know More An irrational or disproportionate fear, aversion, or resistance to advanced technologies, technological change, and digital innovation. Manifests as psychological and physiological responses ranging from mild anxiety to severe distress when interacting with or contemplating technological systems. Often characterised by: Cognitive resistance to learning new technological skills Physical symptoms when forced to use technology Avoidance behaviours toward digital tools and platforms Distinguished from rational technology criticism by its emotional rather than analytical basis. Particularly relevant in contexts of rapid technological transformation, AI adoption, and digital transformation initiatives. Toolware Know More A category of software tools or AI-driven utilities designed to assist in specific tasks within the software development lifecycle, often used in a decentralized or uncoordinated manner across development teams. In computing, a collection of integrated or standalone applications and agents that support development, testing, or deployment processes, sometimes leading to workflow sprawl if not governed by a unified framework. Token Economics Know More The cost-performance analysis framework governing enterprise LLM deployment decisions based on the computational expense of processing input and output tokens, measured in tokens per dollar and tokens per second. Token economics encompasses tradeoffs between model size, context window length, inference latency, throughput requirements, and operational budgets that determine architectural choices between proprietary APIs versus self-hosted models. This economic calculus has emerged as a primary driver of SLM adoption, prompt optimization practices, and hybrid deployment strategies as organizations confront the reality that serving costs often exceed training expenses. Transformer Model Know More A neural network architecture introduced in the 2017 Google paper "Attention Is All You Need" that uses self-attention mechanisms to process sequential data. Transformers can determine relationships between elements in a sequence without the need for recurrent connections, enabling more efficient parallel processing. Transformer models consist of encoder and decoder components working together with an attention mechanism that weighs the importance of different elements in the input sequence. This architecture has proven remarkably versatile, powering advances in natural language processing, computer vision, and multimodal AI. Transformers form the foundation of large language models (LLMs) like ChatGPT and have enabled significant breakthroughs in AI's ability to understand and generate human-like perceivable content. Their ability to process all elements of a sequence in parallel (rather than sequentially) has dramatically improved training efficiency compared to earlier architectures. Go to IndoPacific.App Wish to read more about any term defined in our Glossary? Go to indopacific.app and search our publications. vwxyz WANA Know More WANA is an official term used by the Government of India to refer to the West Asia and North Africa region. The Ministry of External Affairs (MEA) of India has a dedicated WANA Division that handles "all matters relating to Algeria, Djibouti, Egypt, Israel, Libya, Lebanon, Morocco, Syria, Palestine, Sudan, South Sudan, Somalia, Jordan and Tunisia". India's Ministry of Commerce and Industry also has a WANA Division that deals with India's trade relations with 19 countries in this region, including Bahrain, Kuwait, Oman, Qatar, Iraq, UAE, Saudi Arabia, Egypt, Sudan, Algeria, Morocco, Tunisia, Syria, Jordan, Israel, Lebanon, Yemen, Libya and South Sudan. The term is formally recognized in diplomatic contexts, as evidenced by the first India-France Consultations on West Asia and North Africa Region held on April 12, 2022, where Dr. Pradeep Singh Rajpurohit, Joint Secretary (WANA), MEA, represented India. According to Indian foreign policy framework, the WANA region encompasses all Arab nations as well as South Sudan, with North Africa being considered a direct extension of the Midd le East. WENA Know More An acronym for Western Europe and Northern America, referring to the geographically and economically developed regions that include countries in Western and Central Europe, the United Kingdom, the United States, and Canada. Coined by satirist Karl Sharro, and popularised by Indian journalist Nirmalya Dutta, WENA is used to satirically critique the analytical frameworks often applied to different global regions, particularly in comparison to the Middle East and North Africa (MENA). This term challenges the notion of Western exceptionalism by advocating for the same rigorous scrutiny of social, political, and cultural issues in WENA that is commonly directed towards MENA, promoting a more balanced and equitable examination of diverse regions. Whole-of-Government Response Know More A whole-of-government response under the (proposed) Digital India Act is a coordinated approach to the governance of digital technologies and issues. It involves the participation of all relevant government ministries and agencies, as well as other stakeholders such as industry and academia. The goal of a whole-of-government response is to ensure that digital technologies are used in a way that is beneficial to society, while also mitigating any potential risks. This may involve developing new policies and regulations, investing in research and development, and raising awareness of digital issues. This was discussed in Deciphering Regulative Methods for Generative AI, VLiGTA-TR-002 (2023) . Zero Knowledge Systems Know More Zero-knowledge systems (ZKSs) are cryptographic protocols that allow one party (the prover) to prove to another party (the verifier) that a statement is true, without revealing any information about the statement itself or how it is proven. ZKSs are based on the idea that it is possible to prove the possession of knowledge without revealing the knowledge itself. This was discussed in Reinventing & Regulating Policy Use Cases of Web3 for India, VLiGTA-TR-004 (2023). Zero Knowledge Taxes Know More Zero-knowledge taxes (ZKTs) are a hypothetical type of tax that could be implemented using ZKSs. ZKTs would allow taxpayers to prove to the government that they have paid their taxes without revealing their income or other financial information. This was discussed in Reinventing & Regulating Policy Use Cases of Web3 for India, VLiGTA-TR-004 (2023). Wish to read more about any term defined in our Glossary? Go to indopacific.app and search our publications. Go to IndoPacific.App terms of use This glossary of terms is provided as a free resource for educational and informational purposes only. By using this glossary developed by Indic Pacific Legal Research LLP (referred to as 'The Firm'), you agree to the following terms of use: You may use the glossary for personal and non-commercial purposes only. If you use any content from the glossary of terms on this website in your own work, you must properly attribute the source. This means including a link to this website and citing the title of the glossary. Here is a sample format to cite this glossary (we have used the OSCOLA citation format as an example): Indic Pacific Legal Research LLP, 'TechinData.in Explainers' (Indic Pacific Legal Research , 2023) You are not authorised to reproduce, distribute, or modify the glossary without the express written permission of a representative of Indic Pacific Legal Research. The Firm makes no representations or warranties about the accuracy or completeness of the glossary. The glossary is provided on an "as is" basis and the Firm disclaims all liability for any errors or omissions in the glossary. You agree to indemnify and hold the Firm harmless from any claims or damages arising out of your use of the glossary. If you have any questions or concerns about these terms of use, please contact us at global@indicpacific.com Law & Policy 101 This section offers free & basic explainers on certain concepts of Law and Policy for general understanding.

lexploit Date of Addition 25 May 2026 The term implies itself to be a cybersecurity exploit in which a document is intentionally manipulated, typically at the font-rendering layer to trick an Artificial Intelligence (AI) system or a Large Language Model (LLM) into reading text that is completely different from what is visible to a human reader. Key Characteristics: Mechanism: Unlike "hidden text" or "white ink" tricks, a lexploit operates at the foundational rendering level of the document ( e.g., utilizing custom fonts like noroboto.tff ) . Distinct from Prompt Injection: While prompt injection manipulates the instructions an AI is given, a lexploit manipulates the actual source data the AI perceives during ingestion. Use Cases Offensive ("Weaponized Hallucination"): Deceiving an AI during automated document review. For example, formatting a contract so a human reads it as being "governed by Maryland law," while an AI conducting M&A due diligence misreads it as "governed by Delaware law." Defensive (Anti-Scraping): Protecting intellectual property by rendering documents invisible or garbled to automated AI ingestion pipelines and scraping agents, while keeping the content perfectly legible to human readers. Attributions & Credits Concept & Terminology: Coined and demonstrated by the team at LegalQuants , an organization focused on ethical hacking, enterprise security, and cyber defense in the legal industry. Demonstration & Development: Articulated by LegalQuants co-founders Raymond Sun (who demonstrated its defensive anti-scraping applications) and Jamie Tso (who demonstrated its offensive M&A applications). Technical Execution: The underlying proof-of-concept font ( noroboto.tff ) was developed by the LegalQuants Red Team: Drew Miller, Iris Ng, Andrius Petrenas, and Aleks Valkov . Related Long-form Insights on IndoPacific.App NIST Adversarial Machine Learning Taxonomies: Decoded, IPLR-IG-016 Learn More Previous Term Next Term The Indic Pacific Glossary The Complete Glossary terms of use This glossary of terms is provided as a free resource for educational and informational purposes only. By using this glossary developed by Indic Pacific Legal Research LLP (referred to as 'The Firm'), you agree to the following terms of use: You may use the glossary for personal and non-commercial purposes only. If you use any content from the glossary of terms on this website in your own work, you must properly attribute the source. This means including a link to this website and citing the title of the glossary. Here is a sample format to cite this glossary (we have used the OSCOLA citation format as an example): Indic Pacific Legal Research LLP, 'TechinData.in Explainers' (Indic Pacific Legal Research , 2023) You are not authorised to reproduce, distribute, or modify the glossary without the express written permission of a representative of Indic Pacific Legal Research. The Firm makes no representations or warranties about the accuracy or completeness of the glossary. The glossary is provided on an "as is" basis and the Firm disclaims all liability for any errors or omissions in the glossary. You agree to indemnify and hold the Firm harmless from any claims or damages arising out of your use of the glossary. If you have any questions or concerns about these terms of use, please contact us at global@indicpacific.com

Madras High Court judgment dated January 5, 2026 in CMA(PT) No. 9 of 2024 dismissing Flipkart Internet Private Ltd's appeal against Patent Office order upholding validity of Voicemonk Inc's Indian Patent IN 312437 for voice-AI virtual agent system enabling conversational commerce through natural language processing and action correlation technology; establishes patentability standards for AI-driven virtual assistants under Section 3(k) Patents Act 1970; applies Seven Stambhas novelty assessment framework from Lava International v Ericsson precedent; confirms voice-based conversational AI systems with sequential hierarchical lateral correlation logic not barred as mere computer programme per se; binding precedent on AI patent validity post-grant opposition appeals under Section 117A Patents Act. Flipkart Internet Private Ltd v. Joint Controller of Patents and Designs & Voicemonk Inc., CMA(PT) No. 9 of 2024, Madras High Court, Order dated January 5, 2026 Madras High Court judgment dated January 5, 2026 in CMA(PT) No. 9 of 2024 dismissing Flipkart Internet Private Ltd's appeal against Patent Office order upholding validity of Voicemonk Inc's Indian Patent IN 312437 for voice-AI virtual agent system enabling conversational commerce through natural language processing and action correlation technology; establishes patentability standards for AI-driven virtual assistants under Section 3(k) Patents Act 1970; applies Seven Stambhas novelty assessment framework from Lava International v Ericsson precedent; confirms voice-based conversational AI systems with sequential hierarchical lateral correlation logic not barred as mere computer programme per se; binding precedent on AI patent validity post-grant opposition appeals under Section 117A Patents Act. Previous Next The AIACT.IN India AI Regulation Tracker This is a simple regulatory tracker consisting all information on how India is regulating artificial intelligence as a technology, inspired from a seminal paper authored by Abhivardhan and Deepanshu Singh for the Forum of Federations, Canada, entitled, "Government with Algorithms: Managing AI in India’s Federal System – Number 70 ". We have also included case laws along with regulatory / governance documents, and avoided adding any industry documents or policy papers which do not reflect any direct or implicit legal impact. January 2026 Read the Document Issuing Authority Madras High Court Type of Legal / Policy Document Judicial Pronouncements - National Court Precedents Status Enacted Regulatory Stage Regulatory Binding Value Legally binding instruments enforceable before courts AIACT. Regulation Visualiser Find more sources Related Long-form Insights on IndoPacific.App Averting Framework Fatigue in AI Governance [IPLR-IG-013] Learn More Indo-Pacific Research Ethics Framework on Artificial Intelligence Use [IPac AI] Learn More The Global AI Inventorship Handbook, First Edition [RHB-AI-INVENT-001-2025] Learn More Artificial Intelligence, Market Power and India in a Multipolar World Learn More Related draft AI Law Provisions of aiact.in Section 3 – Classification of Artificial Intelligence Section 3 – Classification of Artificial Intelligence Section 14 – Model Standards on Knowledge Management Section 14 – Model Standards on Knowledge Management Section 15 – Guidance Principles for AI-related Agreements Section 15 – Guidance Principles for AI-related Agreements Section 21 – Intellectual Property Protections Section 21 – Intellectual Property Protections

Manipur High Court May 2024 judgment using ChatGPT for legal research on VDF service rules resulting in petitioner's reinstatement. Md Zakir Hussain v. State of Manipur, W.P. (C) No. 1080 of 2023 (Manipur High Court, May 23, 2024) Manipur High Court May 2024 judgment using ChatGPT for legal research on VDF service rules resulting in petitioner's reinstatement. Previous Next The AIACT.IN India AI Regulation Tracker This is a simple regulatory tracker consisting all information on how India is regulating artificial intelligence as a technology, inspired from a seminal paper authored by Abhivardhan and Deepanshu Singh for the Forum of Federations, Canada, entitled, "Government with Algorithms: Managing AI in India’s Federal System – Number 70 ". We have also included case laws along with regulatory / governance documents, and avoided adding any industry documents or policy papers which do not reflect any direct or implicit legal impact. May 2024 Read the Document Issuing Authority Manipur High Court Type of Legal / Policy Document Judicial Pronouncements - National Court Precedents Status Enacted Regulatory Stage Miscellaneous Binding Value Legally binding instruments enforceable before courts AIACT. Regulation Visualiser Find more sources Related Long-form Insights on IndoPacific.App Reimaging and Restructuring MeiTY for India [IPLR-IG-007] Learn More Averting Framework Fatigue in AI Governance [IPLR-IG-013] Learn More Decoding the AI Competency Triad for Public Officials [IPLR-IG-014] Learn More AI Bias & the Overlap of AI Diplomacy and Governance Ethics Dilemmas Learn More Artificial Intelligence, Market Power and India in a Multipolar World Learn More Related draft AI Law Provisions of aiact.in Section 3 – Classification of Artificial Intelligence Section 3 – Classification of Artificial Intelligence Section 7 – Risk-centric Methods of Classification Section 7 – Risk-centric Methods of Classification Section 8 – Prohibition of Unintended Risk AI Systems Section 8 – Prohibition of Unintended Risk AI Systems

Section 5 – Technical Methods of Classification PUBLISHED Previous Next Section 5 – Technical Methods of Classification (1) These methods as designated in clause (ii) of sub-section (1) of Section 3 classify artificial intelligence technologies subject to their scale, inherent purpose, technical features and technical limitations such as – (i) General Purpose Artificial Intelligence Applications with Multiple Stable Use Cases (GPAIS) as described in sub-section (2); (ii) General Purpose Artificial Intelligence Applications with Multiple Short-Run or Unclear Use Cases (GPAIU) as described in sub-section (3); (iii) Specific-Purpose Artificial Intelligence Applications with One or More Associated Standalone Use Cases or Test Cases (SPAI) as described in sub-section (4); (2) General Purpose Artificial Intelligence Systems with Multiple Stable Use Cases (GPAIS) are classified based on a technical method that evaluates the following factors in accordance with relevant sector-specific and sector-neutral industrial standards: (i) Scale: The ability to operate effectively and consistently across a wide range of domains, handling large volumes of data and users. (ii) Inherent Purpose: The capacity to be adapted and applied to multiple well-defined use cases within and across sectors. (iii) Technical Features: Robust and flexible architectures that enable reliable performance on diverse tasks and requirements. (iv) Technical Limitations: Potential challenges in maintaining consistent performance and compliance with sector-specific regulations across the full scope of intended use cases. Illustration An AI system used in healthcare for diagnostics, treatment recommendations, and patient management. This AI consistently performs well in various healthcare settings, adhering to medical standards and providing reliable outcomes. It is characterized by its large scale in handling diverse medical data and serving multiple institutions, its inherent purpose of assisting healthcare professionals in decision-making and care improvement, robust technical architecture and accuracy while adhering to privacy and security standards, and potential limitations in edge cases or rare conditions. (3) General Purpose Artificial Intelligence Systems with Multiple Short-Run or Unclear Use Cases (GPAIU) are classified based on a technical method that evaluates the following factors in accordance with relevant sector-specific and sector-neutral industrial standards: (i) Scale: The ability to address specific short-term needs or exploratory applications within relevant sectors at a medium scale. (ii) Inherent Purpose: Providing targeted solutions for emerging or temporary use cases, with the potential for future adaptation and expansion. (iii) Technical Features: Modular and adaptable architectures enabling rapid development and deployment in response to evolving requirements. (iv) Technical Limitations: Uncertainties regarding long-term viability, scalability, and compliance with changing industry standards and regulations. Illustration An AI system used in experimental smart city projects for traffic management, pollution monitoring, and public safety. Deployed at a medium scale in specific locations for limited durations, its inherent purpose is testing and validating AI feasibility and effectiveness in smart city applications. It features a modular, adaptable technical architecture to accommodate changing requirements and infrastructure integration, but faces potential limitations in scalability, interoperability, and long-term performance due to the experimental nature. (4) Specific-Purpose Artificial Intelligence Systems with One or More Associated Standalone Use Cases or Test Cases (SPAI) are classified based on a technical method that evaluates the following factors: (i) Scale: The ability to address specific, well-defined problems or serve as proof-of-concept implementations at a small scale. (ii) Inherent Purpose: Providing specialized solutions for individual use cases or validating AI technique feasibility in controlled environments. (iii) Technical Features: Focused and optimized architectures tailored to the specific requirements of the standalone use case or test case. (iv) Technical Limitations: Constraints on generalizability, difficulties scaling beyond the initial use case, and challenges ensuring real-world robustness and reliability. Illustration An AI chatbot used by a company for customer service during a product launch. As a small-scale standalone application, its inherent purpose is providing automated support for a specific product or service. It employs a focused, optimized technical architecture for handling product-related queries and interactions, but faces limitations in handling queries outside the predefined scope or adapting to new products without significant modifications. Related Indian AI Regulation Sources

Section 10 – Composition and Functions of the Council PUBLISHED Previous Next Section 10 - Composition and Functions of the Council (1) With effect from the date notified by the Central Government, there shall be established the Indian Artificial Intelligence Council (IAIC), a statutory body for the purposes of this Act. (2) The IAIC shall be an autonomous body corporate with perpetual succession, a common seal, and the power to acquire, hold and transfer property, both movable and immovable, and to contract and be contracted, and sue or be sued by its name. (3) The IAIC shall coordinate and oversee the development, deployment, and governance of artificial intelligence systems across all government bodies, ministries, departments, and regulatory authorities, adopting a whole-of-government approach. (4) The headquarters of the IAIC shall be located at the place notified by the Central Government. (5) The IAIC shall consist of a Chairperson and such number of other Members, not exceeding [X], as the Central Government may notify. (6) The Chairperson and Members shall be appointed by the Central Government through a transparent and merit-based selection process, as may be prescribed. (7) The Chairperson and Members shall be individuals of eminence, integrity and standing, possessing specialized knowledge or practical experience in fields relevant to the IAIC’s functions, including but not limited to: (i) Data and artificial intelligence governance, policy and regulation; (ii) Administration or implementation of laws related to consumer protection, digital rights and artificial intelligence and other emerging technologies; (iii) Dispute resolution, particularly technology and data-related disputes; (iv) Information and communication technology, digital economy and disruptive technologies; (v) Law, regulation or techno-regulation focused on artificial intelligence, data protection and related domains; (vi) Any other relevant field deemed beneficial by the Central Government. (8) At least three Members shall be experts in law with demonstrated understanding of legal and regulatory frameworks related to artificial intelligence, data protection and emerging technologies. (9) The IAIC shall have the following functions: (i) Develop and implement policies, guidelines and standards for responsible development, deployment and governance of AI systems in India; (ii) Coordinate and collaborate with relevant ministries, regulatory bodies and stakeholders to ensure harmonised AI governance across sectors; (iii) Establish and maintain the National Registry of AI Use Cases as per Section 12; (iv) Administer the certification scheme for AI systems as specified in Section 11; (v) Develop and promote the National AI Ethics Code as outlined in Section 13; (vi) Facilitate stakeholder consultations, public discourse and awareness on societal implications of AI; (vii) Promote research, development and innovation in AI with a focus on responsibility and ethics; (viii) Engage with international AI regulatory bodies, standard-setting organizations, and global AI safety initiatives to promote knowledge exchange and align India’s AI governance framework with global best practices. This includes: (a) Developing bilateral and multilateral agreements to support collaborative research, data sharing, and risk management. (b) Participating in international AI safety and ethics dialogues to shape global AI norms. (c) Coordinating on cross-border data flow standards and AI certification criteria to ensure seamless compliance for international AI applications in India. (ix) Take regulatory actions to ensure compliance with the policies, standards, and guidelines issued by the IAIC under this Act, which may include: (a) Issuing show-cause notices requiring non-compliant entities to explain the reasons for non-compliance and outline corrective measures within a specified timeline; (b) Imposing monetary penalties based on the severity of non-compliance, the risk level involved, and the potential impact on individuals, businesses, or society, with penalties being commensurate with the financial capacity of the non-compliant entity; (c) Suspending or revoking certifications, registrations, or approvals related to non-compliant AI systems, preventing their further development, deployment, or operation until compliance is achieved; (d) Mandating independent audits of the non-compliant entity’s processes at their own cost, with audit reports to be submitted to the IAIC for review and further action; (e) Issuing directives to non-compliant entities to implement specific remedial measures within a defined timeline, such as enhancing data quality controls, improving governance frameworks, or strengthening decision-making procedures; (f) In cases of persistent or egregious non-compliance, recommending the temporary or permanent suspension of the non-compliant entity’s AI-related operations, subject to due process and the principles of natural justice; (g) Taking any other regulatory action deemed necessary and proportionate to ensure compliance with the prescribed standards and to safeguard the responsible development, deployment, and use of AI systems. (x) Advise the Central Government on matters related to AI policy, regulation and governance, and recommend legislative or regulatory changes as necessary; (xi) Perform any other functions necessary to achieve the objectives of this Act or as assigned by the Central Government. (10)The IAIC may constitute advisory committees, expert groups or task forces as deemed necessary to assist in its functions. (11)The IAIC shall endeavour to function as a digital office to the extent practicable, conducting proceedings, filings, hearings and pronouncements through digital means as per applicable laws. Related Indian AI Regulation Sources Report on AI Governance Guidelines Development January 2025 Framework for Responsible and Ethical Enablement of Artificial Intelligence (FREE-AI) Committee Report August 2025 India AI Governance Guidelines: Enabling Safe and Trusted AI Innovation November 2025 Digital Personal Data Protection Rules, 2025 November 2025 Working Paper on Generative AI and Copyright (Part 1): "One Nation One License One Payment" December 2025 Democratising Access to AI Infrastructure (White Paper, Version 3.0) December 2025

Model Collapse Date of Addition 17 October 2025 A degenerative phenomenon where AI models trained on recursively generated synthetic data progressively lose diversity, accuracy, and quality over successive training iterations, ultimately producing increasingly homogeneous and corrupted outputs. This feedback loop occurs when models consume their own generated content or outputs from similar models as training data, causing statistical distributions to narrow and tail events to disappear from learned representations. Model collapse poses existential risks to the long-term viability of AI systems as synthetic content proliferates across the internet, contaminating datasets used for future model training. Related Long-form Insights on IndoPacific.App NIST Adversarial Machine Learning Taxonomies: Decoded, IPLR-IG-016 Learn More Previous Term Next Term The Indic Pacific Glossary The Complete Glossary terms of use This glossary of terms is provided as a free resource for educational and informational purposes only. By using this glossary developed by Indic Pacific Legal Research LLP (referred to as 'The Firm'), you agree to the following terms of use: You may use the glossary for personal and non-commercial purposes only. If you use any content from the glossary of terms on this website in your own work, you must properly attribute the source. This means including a link to this website and citing the title of the glossary. Here is a sample format to cite this glossary (we have used the OSCOLA citation format as an example): Indic Pacific Legal Research LLP, 'TechinData.in Explainers' (Indic Pacific Legal Research , 2023) You are not authorised to reproduce, distribute, or modify the glossary without the express written permission of a representative of Indic Pacific Legal Research. The Firm makes no representations or warranties about the accuracy or completeness of the glossary. The glossary is provided on an "as is" basis and the Firm disclaims all liability for any errors or omissions in the glossary. You agree to indemnify and hold the Firm harmless from any claims or damages arising out of your use of the glossary. If you have any questions or concerns about these terms of use, please contact us at global@indicpacific.com

Issued in October 2025 by the Assistant Director (MSU) with approval of the Director General, ESIC. This circular directs all ESIC employees to refrain from using open or external AI tools such as ChatGPT, DeepSeek, Copilot, Midjourney, DALL-E, Jasper, and Adobe Firefly for official noting, drafting, or data analysis. The circular cites significant risks to data security and leakage of sensitive information related to stakeholders. Circular on Use of Open/External Artificial Intelligence (AI) Tools for Official Work, File No. Z-11/12/1/Misc.Matter/2024-MSU Issued in October 2025 by the Assistant Director (MSU) with approval of the Director General, ESIC. This circular directs all ESIC employees to refrain from using open or external AI tools such as ChatGPT, DeepSeek, Copilot, Midjourney, DALL-E, Jasper, and Adobe Firefly for official noting, drafting, or data analysis. The circular cites significant risks to data security and leakage of sensitive information related to stakeholders. Previous Next The AIACT.IN India AI Regulation Tracker This is a simple regulatory tracker consisting all information on how India is regulating artificial intelligence as a technology, inspired from a seminal paper authored by Abhivardhan and Deepanshu Singh for the Forum of Federations, Canada, entitled, "Government with Algorithms: Managing AI in India’s Federal System – Number 70 ". We have also included case laws along with regulatory / governance documents, and avoided adding any industry documents or policy papers which do not reflect any direct or implicit legal impact. October 2025 Read the Document Issuing Authority Employees' State Insurance Corporation (ESIC), Ministry of Labour and Employment Type of Legal / Policy Document Executive Instruments - Administrative Decisions Status In Force Regulatory Stage Regulatory Binding Value Legally binding instruments enforceable before courts AIACT. Regulation Visualiser Find more sources Related Long-form Insights on IndoPacific.App Reimaging and Restructuring MeiTY for India [IPLR-IG-007] Learn More Averting Framework Fatigue in AI Governance [IPLR-IG-013] Learn More Decoding the AI Competency Triad for Public Officials [IPLR-IG-014] Learn More AI Bias & the Overlap of AI Diplomacy and Governance Ethics Dilemmas Learn More Artificial Intelligence, Market Power and India in a Multipolar World Learn More Related draft AI Law Provisions of aiact.in Section 15 – Guidance Principles for AI-related Agreements Section 15 – Guidance Principles for AI-related Agreements Section 16 – Guidance Principles for AI-related Corporate Governance Section 16 – Guidance Principles for AI-related Corporate Governance

Delhi High Court October 2025 judgment protecting Bollywood actor Hrithik Roshan's personality rights against AI-generated deepfakes, morphed content, and unauthorized merchandise exploitation while exempting non-commercial fan pages from blanket takedown orders. Significantly, the court balanced personality rights protection with fan expression by declining to order immediate takedown of fan pages, noting they use the actor's image for non-commercial purposes. Hrithik Roshan v. Ashok Kumar/John Doe & Ors., CS(COMM) 1107/2025, Delhi High Court, Order dated October 15, 2025 Delhi High Court October 2025 judgment protecting Bollywood actor Hrithik Roshan's personality rights against AI-generated deepfakes, morphed content, and unauthorized merchandise exploitation while exempting non-commercial fan pages from blanket takedown orders. Significantly, the court balanced personality rights protection with fan expression by declining to order immediate takedown of fan pages, noting they use the actor's image for non-commercial purposes. Previous Next The AIACT.IN India AI Regulation Tracker This is a simple regulatory tracker consisting all information on how India is regulating artificial intelligence as a technology, inspired from a seminal paper authored by Abhivardhan and Deepanshu Singh for the Forum of Federations, Canada, entitled, "Government with Algorithms: Managing AI in India’s Federal System – Number 70 ". We have also included case laws along with regulatory / governance documents, and avoided adding any industry documents or policy papers which do not reflect any direct or implicit legal impact. October 2025 Read the Document Issuing Authority Delhi High Court Type of Legal / Policy Document Judicial Pronouncements - National Court Precedents Status In Force Regulatory Stage Regulatory Binding Value Legally binding instruments enforceable before courts AIACT. Regulation Visualiser Find more sources Related Long-form Insights on IndoPacific.App Regularizing Artificial Intelligence Ethics in the Indo-Pacific [GLA-TR-002] Learn More Impact-Based Legal Problems around Generative AI in Publishing, IPLR-IG-010 Learn More Indo-Pacific Research Ethics Framework on Artificial Intelligence Use [IPac AI] Learn More The Global AI Inventorship Handbook, First Edition [RHB-AI-INVENT-001-2025] Learn More Related draft AI Law Provisions of aiact.in Section 21 – Intellectual Property Protections Section 21 – Intellectual Property Protections Section 23 – Content Provenance and Identification Section 23 – Content Provenance and Identification

Section 10-A – Composition and Functions of the Institute PUBLISHED Previous Next Section 10-A – Composition and Functions of the Institute (1) With effect from the date notified by the Central Government, there shall be established the Indian Artificial Intelligence Safety Institute (AISI), a statutory body for the purposes of this Act. (2) The Indian Artificial Intelligence Safety Institute (AISI) shall be established as an autonomous body corporate with perpetual succession, a common seal, and the power to acquire, hold and transfer property, both movable and immovable, and to contract and be contracted, and sue or be sued by its name. (3) The Governing Body of the Indian Artificial Intelligence Safety Institute shall consist of the following members: (i) A Director General of AI Safety, with at least 15 years of experience in artificial intelligence research, who shall serve as the Chief Executive Officer of AISI. (ii) One representative from the Ministry of Electronics and Information Technology (MeitY), not below the rank of Joint Secretary. (iii) One representative from the Ministry of Science and Technology (DST), not below the rank of Joint Secretary. (iv) One representative from the Ministry of Defence, not below the rank of Joint Secretary. (v) One representative from the Ministry of Communications, not below the rank of Joint Secretary. (vi) One representative from NITI Aayog, not below the rank of Joint Secretary. (vii)One representative from the Committee for AI Centers of Excellence (CoEs) as an ex-officio member (viii) One Representative from the Committee for Technical Institutions in Critical AI Research as an ex-officio member (ix) One Representative from the Committee on AI Ethics and Safety as an ex-officio member (4) In addition to the Governing Body, AISI shall include the following ex-officio members: (i) The Principal Scientific Advisor to the Government of India, or their nominee. (ii) One member from the Prime Minister’s Economic Advisory Council . (iii) One representative, being a government official or expert appointed by the Central Government, responsible for coordinating with global AI safety institutes to ensure knowledge exchange and collaboration on emerging risks and best practices. (5) The AISI shall establish specialized committees as deemed necessary for fulfilling its mandate. These committees shall include but are not limited to: (i) Committee for AI Centers of Excellence (CoEs) : This committee shall represent all AI-related Centers of Excellence across India. (ii) Committee for Technical Institutions in Critical AI Research : This committee shall coordinate with technical institutions engaged in critical research on AI systems. (iii) Committee on AI Ethics and Safety : This committee shall guide AISI on ethical principles governing AI systems. (6) The AISI shall undertake the following functions under this Act: (i) Develop protocols for risk assessment, monitoring, and mitigation concerning high-risk AI applications, particularly in strategic sectors such as healthcare, defence, finance, and public administration. (ii) Formulate and establish safety standards for high-risk AI applications for the IAIC. These standards shall be aligned with national security priorities and international norms governing AI safety. (iii) Conduct annual audits of high-risk AI systems deployed across various sectors. The findings from these audits shall be reported to IAIC for further action or policy formulation. (iv) Undertake research initiatives focused on identifying emerging risks associated with new developments in artificial intelligence. Such research shall be conducted in partnership with IAIC, academic institutions, technical bodies and centres of excellence (CoEs), and international organizations dedicated to AI safety. (v) Submit an annual report to the Central Government and IAIC, detailing safety incidents, audit findings, and research advancements. (7) AISI may engage in international partnerships and dialogues, contributing to India’s leadership in responsible AI governance. Related Indian AI Regulation Sources

Section 17 – Post-Deployment Monitoring of High-Risk AI Systems PUBLISHED Previous Next Section 17 - Post-Deployment Monitoring of High-Risk AI Systems (1) High-risk AI systems as classified in the sub-section (4) of Section 7 shall be subject to ongoing monitoring and evaluation throughout their lifecycle to ensure their safety, security, reliability, transparency and accountability. (2) The post-deployment monitoring shall be conducted by the providers, deployers, or users of the high-risk AI systems, as appropriate, in accordance with the guidelines established by the IAIC. (3) The IAIC shall develop and establish comprehensive guidelines for the post-deployment monitoring of high-risk AI systems, which may include, but not be limited to, the following: (i) Identification and assessment of potential risks, which includes: (a) performance deviations, (b) malfunctions, (c) unintended consequences, (d) security vulnerabilities, and (e) data breaches; (ii) Evaluation of the effectiveness of risk mitigation measures and implementation of necessary updates, corrections, or remedial actions; (iii) Continuous improvement of the AI system’s performance, reliability, and trustworthiness based on real-world feedback and evolving best practices; and (iv) Regular reporting to the IAIC on the findings and actions taken as a result of the post-deployment monitoring, including any incidents, malfunctions, or adverse impacts identified, and the measures implemented to address them. (4) The post-deployment monitoring facilitated by the IAIC shall involve collaboration and coordination among providers, deployers, users, and sector-specific regulatory authorities, to ensure a comprehensive and inclusive approach to AI system oversight. Related Indian AI Regulation Sources Advisory on AI Intermediaries and Platforms March 2024 Report on AI Governance Guidelines Development January 2025 India AI Governance Guidelines: Enabling Safe and Trusted AI Innovation November 2025 Digital Personal Data Protection Rules, 2025 November 2025 Working Paper on Generative AI and Copyright (Part 1): "One Nation One License One Payment" December 2025 Democratising Access to AI Infrastructure (White Paper, Version 3.0) December 2025