SDG 9: Industry, Innovation and Infrastructure | SDG 16: Peace, Justice and Strong Institutions | SDG 17: Partnerships for the Goals
Office of the Principal Scientific Adviser (PSA) | MeitY | Bureau of Indian Standards (BIS) | NITI Aayog
The Office of the Principal Scientific Adviser (PSA) has released a white paper titled Strengthening AI Governance Through Techno-Legal Framework. The paper proposes a shift from traditional “command-and-control” regulations to a techno-legal approach, which integrates legal instruments with technical enforcement mechanisms embedded directly into AI architecture by design. This model aims to resolve the “pacing problem”—where technology outsteps traditional law—by encoding legal obligations like privacy, fairness, and safety into the system’s technical artifacts at the design stage.
Lifecycle-Based Safety and Trust The framework mandates that “Safe and Trusted AI” attributes must be complied with across all five stages of the AI lifecycle:
Data Collection & Protection: Utilizing Privacy-Enhancing Technologies (PETs) like differential privacy and synthetic data to protect personal and proprietary info.
Model Training & Assessment: Implementing “Machine Unlearning” to allow models to “forget” specific data influences, supporting the legal right to erasure.
Safe AI Inference: Deploying “Responsible AI Firewalls” to mitigate real-time risks like hallucinations, deepfakes, and prompt injections.
Trusted Agents: Establishing kill switches, behavior logging, and agentic guardrails for autonomous systems to prevent large-scale automated failures.
Institutional Architecture for Implementation The white paper outlines a “whole-of-government” institutional mechanism to operationalize this framework:
AI Governance Group (AIGG): Chaired by the PSA, this body will coordinate between ministries and regulators to establish uniform, sector-agnostic standards.
Technology and Policy Expert Committee (TPEC): A multidisciplinary unit within MeitY to bridge the gap between emerging AI capabilities and regulatory needs.
AI Safety Institute (AISI): The central hub for evaluating and testing high-risk systems and developing open-source safety toolkits like Nishpaksh for fairness audits.
National AI Incident Database: A centralized record for tracking safety failures, biased outcomes, and security breaches to inform data-driven audits.
What is “Machine Unlearning” in the context of India’s techno-legal framework? Machine Unlearning is a technical capability that allows a trained AI model to “forget” or remove the influence of specific data points without requiring a full retraining of the model. In a techno-legal context, it is a critical tool for operationalizing the “Right to be Forgotten” under the DPDP Act 2023, enabling organizations to comply with data erasure or consent revocation requests in a scalable, cost-effective manner.
Policy Relevance
The techno-legal framework is designed to position India as a global leader in “Responsible AI by Design,” particularly for population-scale deployments.
Leveraging DPI and DEPA: The framework integrates with India’s Digital Public Infrastructure (DPI) and Data Empowerment and Protection Architecture (DEPA) to enable consent-driven, purpose-limited data sharing for AI training.
Addressing Demographic Diversity: By mandating India-specific evaluations, the policy ensures AI systems are tested for local accents, multilingual usage, and skin-tone sensitivity, preventing “Western-centric” biases from harming Indian citizens.
Balancing Rights and Performance: The paper notes a structural tension between privacy and model performance; it recommends “impact-aware data withdrawal” to ensure that large-scale data erasure does not lead to the demographic exclusion of underrepresented groups.
SME and Startup Enablement: By utilizing DPI for low-cost compliance checks, the framework aims to reduce the regulatory burden on smaller firms while maintaining high safety standards.
Follow the full paper here: STRENGTHENING AI GOVERNANCE THROUGH TECHNO-LEGAL FRAMEWORK