Centre weighs curbs on AI autonomy, consent rules for deepfakes under new law

📌 Summary:

• The Ministry of Electronics and IT (MeitY) is exploring a standalone law to regulate artificial intelligence, rather than placing it under the IT Act, 2000; its cyber laws division has been asked to analyse gaps in the IT Act and its rules
• A consent-based framework is being considered for "synthetically-generated content" (digital media made or altered by AI), driven by the rapid proliferation of deepfakes created without the consent of the individuals whose images/videos are used
• For agentic AI — autonomous systems that plan multi-step actions, use external tools and adjust strategy with minimal human supervision — the government is weighing how much autonomy to allow and whether such agents may retain and reuse data
• Regulatory sandboxes are proposed for high-risk applications, especially in finance and public services, with RBI and SEBI to be consulted
• Platform liability & safe harbour: unlike social media (which enjoys "safe harbour" immunity for user-generated content), AI models editorialise responses that are hard to trace to a source, so that logic does not translate directly; two private-sector legal experts have been asked to submit draft liability frameworks
• Gap with existing law: while the Digital Personal Data Protection (DPDP) Act, 2023 covers broad privacy, questions of consent and data ownership in the AI age need further examination

🎯 UPSC Relevance: GS3 Science & Technology / GS2 Governance — regulation of emerging technologies (AI), data protection, deepfakes, and balancing innovation with accountability.

📝 Prelims Facts:

• Nodal ministry: MeitY (Ministry of Electronics and Information Technology); cyber laws division tasked with the review
• "Safe harbour" is legal immunity for platforms from liability for user-generated content
• Digital Personal Data Protection (DPDP) Act was enacted in 2023

🔑 Key Term: Agentic AI — autonomous AI systems that, given a high-level goal, independently plan and execute multi-step actions using external tools, with minimal human supervision.