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

📌 Summary:

• MeitY is exploring a standalone AI law with three planks: a consent-based framework for synthetically-generated content, curbs on agentic-AI autonomy, and regulatory sandboxes for high-risk applications
• The law is likely to be separate from the IT Act, 2000; the Ministry's cyber-laws division is analysing gaps in the IT Act and its rules
• For agentic AI, the government will weigh how much autonomy such agents should have and whether they may retain and reuse data (agentic AI independently plans multi-step actions and uses external tools with minimal human oversight)
• RBI and SEBI are to be consulted to set up a regulatory sandbox for AI in high-impact areas like finance and public services
• Liability is a core unresolved issue: unlike social-media "safe harbour", AI models are trained on internet content and "editorialise" responses that are hard to trace to a source; two private legal experts have been asked to draft liability frameworks
• Deepfake concern: most deepfakes use individuals' images/videos without consent; a consent-based framework for synthetic media is seen as a key requirement
• The DPDP Act, 2023 covers broad privacy but consent/data-ownership questions in the AI age "merit further examination"; notably, data voluntarily made public is exempted from the Act's protections, leaving scraped public social-media images largely unguarded
• Policy shift: the government earlier said India did not need an AI law and had enacted only piecemeal measures (e.g., April rules for social-media platforms)

🎯 UPSC Relevance: GS2 Governance — regulating emerging technology, data protection, platform liability and consent; balancing innovation with citizen rights.

📝 Prelims Facts:

• MeitY = Ministry of Electronics and Information Technology; likely to propose a standalone AI law (not under the IT Act, 2000)
• Agentic AI autonomously plans and executes multi-step tasks using external tools
• The DPDP Act, 2023 exempts personal data voluntarily made public from its protections
• "Safe harbour" gives platforms legal immunity for user-generated content

🔑 Key Term: Safe harbour — legal immunity protecting intermediaries/platforms from liability for content posted by users, a principle difficult to apply to generative-AI outputs.