India's AI reckoning: MeitY confronts Grok over deepfakes as DeepSeek V4 faces distillation charges

Two AI controversies converged on India this week, each exposing a different failure mode of the technology industry's headlong rush toward capability without accountability.

Grok: When Image Generation Becomes a Weapon

India's Ministry of Electronics and Information Technology issued a 72-hour compliance notice to X (formerly Twitter), demanding the platform address the misuse of its AI chatbot Grok for generating non-consensual, morphed, and sexually explicit images. The notice — issued under the IT Act 2000 and the Digital Personal Data Protection Act 2023 — came after reports that users were using Grok's image generation capabilities to create deepfake pornography of real individuals, including public figures and minors.

Grok 4.20, Elon Musk's latest AI model released in public beta this month, has significantly fewer content guardrails than competitors like ChatGPT, Claude, or Gemini. While Musk has positioned this as a commitment to "free speech AI," critics argue it has made Grok the preferred tool for generating non-consensual intimate imagery.

India is not alone in its concern. Grok is currently under investigation or regulatory action in:

• France: CNIL investigation into non-consensual image generation
• California: State AG inquiry into minors' safety
• United Kingdom: OFCOM action under the Online Safety Act
• Brazil: Federal police investigation
• Ireland: DPC probe into GDPR violations

DeepSeek V4 and the Distillation Accusations

Simultaneously, Chinese AI startup DeepSeek is preparing to launch its next-generation model, DeepSeek V4, around the Lunar New Year. Internal benchmarks suggest V4 could outperform Claude and GPT in coding tasks — a claim that, if verified, would represent a significant shift in the competitive landscape.

But OpenAI has cast a shadow over the launch by publicly accusing DeepSeek of "distilling" its models — a technique where an adversary uses automated scripts to query a deployed model thousands of times, collecting input-output pairs that can then be used to train a competing model. OpenAI alleges that DeepSeek employees bypassed access controls on its API to conduct large-scale data extraction.

The accusation matters beyond corporate rivalry. If model distillation becomes normalised, the economic incentive to invest billions in original research collapses. Why spend $100 million training a frontier model when a competitor can replicate its capabilities for $1 million through clever querying?

India's Position

India finds itself at the intersection of both controversies. As the world's largest democracy and a country with over 800 million internet users, it is both a massive market for AI companies and a jurisdiction with the regulatory will to act. The AI Impact Summit, running concurrently in New Delhi, provided the diplomatic backdrop for these enforcement actions — signalling that India intends to be not just a consumer of AI but an active regulator of its harms.