The CDSCO Guidance Document on Medical Device Software (CDSCO/MD/GD/MDSW/01/2026) functions as an interpretive framework clarifying how software—specifically Software as a Medical Device (SaMD) and embedded code—is regulated under India’s Medical Devices Rules, 2017 (MDR-2017). For applications utilising AI/ML algorithms, the CDSCO enforces specific technical evaluation components across the engineering lifecycle, such as (1) mnufacturers must document the exact software engineering lineage, which includes separating locked (static) algorithms from adaptive ones; (2) Submissions must explicitly detail the datasets used for training, tuning, and validation; (3) Because AI models are prone to data drift (where changes in real-world clinical practices or equipment cause performance to degrade over time), the guidance mandates continuous real-world performance monitoring. Manufacturers must track metrics like sensitivity, specificity, and Area Under the Curve (AUC) against real-world data pipelines; and (4) Developers must technically validate how the AI model arrives at its conclusions, detailing user-interface mechanisms that inform clinicians of data boundaries, confidence scores, or potential false positives/negatives.