India considers dedicated AI law

In a major shift in digital policy, the Union Ministry of Electronics and Information Technology (MeitY) has officially commenced discussions to introduce a dedicated, standalone legal framework to regulate artificial intelligence (AI).

The announcement, made by MeitY Secretary S. Krishnan on July 3, 2026, represents a sharp departure from the government’s long-standing position that existing cyber statutes and data protection laws were sufficient to police emerging software harms. Electronics and IT Minister Ashwini Vaishnaw simultaneously noted that because the foundational Information Technology Act was enacted all the way back in 2000—long before the advent of generative AI—the tech has outgrown the current safety net, necessitating a comprehensive, independent law.

1. The Shifting Stance: Moving Past the 2026 IT Rules

For years, India approached AI through reactive “soft guidelines” and platform advisories. That changed incrementally in February 2026, when MeitY enacted the Information Technology Amendment Rules, 2026, specifically to establish an emergency buffer against synthetic media:

• The “SGI” Baseline: The February amendment introduced the first legal definition for Synthetically Generated Information (SGI)—classifying any audio, visual, or video material algorithmically altered or created to mimic real-world events or people as high-risk.
• The Aggressive Takedown Mandate: Under these rules, social media intermediaries face strict timelines to remove unlawful SGI once notified, dropping from a standard 36-hour window to a hyper-accelerated 2-hour takedown window for non-consensual deepfakes and intimate imagery.
• Mandatory Provenance: The rule mandated that platforms prominently label AI content and embed permanent, traceable metadata markers into files. Failing to execute these filters strips platforms of their statutory “Safe Harbour” immunity, opening them up to direct criminal liability.

However, MeitY officials now argue that while the 2026 IT Rules successfully tackled immediate social media threats like deepfake impersonations, they are structurally insufficient to govern the broader systemic, economic, and industrial realities of advanced autonomous tech.

2. The Core Focus Areas of the Proposed Legislation

While the legislative text is currently being drafted at the ministerial level, policy working groups confirm that the standalone AI law will pivot away from purely moderating content to executing a comprehensive “techno-legal” framework:

        [ PROPOSED STANDALONE INDIA AI LAW CORRIDORS ]
  
  [ Risk Classification Engine ] ──► Tiers systems from Low Risk to "Unacceptable Risk"
                                              │
                                              ▼
  [ Algorithmic Accountability ] ──► Mandatory pre-lock dataset diversity & bias audits
                                              │
                                              ▼
  [ Sovereign Data Protection  ] ──► Rigid consent architectures for LLM training inputs
                                              │
                                              ▼
  [ Regulatory Oversight Labs  ] ──► Establishes the AI Governance Group (AIGG) & AISI

• Risk-Based Classification: Heavily inspired by global precedents like the EU AI Act, the bill will implement a tiered risk engine. Low-risk tools (like text editing or accessibility enhancements) will face minimal red tape, while high-risk applications (biometric tracking, facial recognition scraping, or algorithmic social credit scoring) will face categorical bans.
• The Guard against “Black Box” Bias: Developers will be legally obligated to provide documented proof of data diversity, ensuring demographic and gender biases are actively scrubbed before foundational weights are locked into consumer-facing models.
• Sectoral Oversight: The law will formally codify dedicated enforcement bodies, including the AI Governance Group (AIGG) and an AI Safety Institute (AISI), mirroring parallel administrative guidelines already taking shape in the Supreme Court (banning AI from core judicial rulings) and the RBI (mandating algorithmic risk audits for automated credit scoring).

3. The Global Balancing Act & Export Relief

The domestic push for regulation lands at a critical moment when India is simultaneously trying to lower international friction for its homegrown engineering sector. Alongside discussions for the new law, MeitY confirmed that export restrictions have successfully been eased on Anthropic’s Mythos model.

AI Frontier Model Tier	Indian Market Accessibility Status (July 2026)	Regulatory / Export Bottleneck
Anthropic Mythos	Eased / Approved	MeitY cleared outbound access barriers, expanding developer usability.
Claude Fable 5 / Mythos 5	Restricted	Deployment remains tightly locked under US government export protocols, restricted exclusively to initial domestic GlassWing corporate partners.

Anthropic has formally stated its intention to expand Fable 5 and Mythos 5 access to entities across 15 allied nations, explicitly including India. However, unlocking those advanced pipelines will require formal clearance from Washington.

By building a specialized, standalone AI law that balances strict citizen safety with a “people-first, responsible innovation” approach, New Delhi is attempting to demonstrate to international partners that it can host advanced foundational systems without sacrificing civil oversight or digital sovereignty.