India Should Write Its Own AI Safety Rules, Not Copy the White House Accord

On September 29, US President Donald Trump and executives from Google, OpenAI, Anthropic, Meta, Nvidia and xAI signed the White House Accord on Super Intelligence. It is a voluntary pact: the companies commit to internal monitoring of their models, dedicated oversight teams, independent external audits and board-level oversight, and they agree to meet regularly to set safety standards.

Indian industry figures responded this week, and on one point they agreed. India should build its own AI safety framework rather than adopt the American one.

What the experts said

• Srikanth Velamakanni, Nasscom chairman and Fractal CEO, called voluntary self-regulation "necessary but woefully insufficient". He wants binding rules on transparency, incident reporting and accountability, and a stronger IndiaAI Safety Institute that tests newly released models without requiring pre-release certification of every one.
• Soumen Mandal of Counterpoint Research backed a hybrid: mandatory rules for defence, healthcare, banking and government use, voluntary standards for fast-moving technical areas.
• Amit Khanna of Grant Thornton Bharat also argued for a "techno-legal" mix, pointing to the synthetic-content provisions in the IT Amendment Rules, 2026 as the binding layer. He noted that India's AI economy is mostly about applications, not foundation models, so reliability across Indian languages matters more than frontier-model controls.
• Ganesh Gopalan of Gnani.ai was the outlier. He favours self-regulation, warns that mandates could favour a few large players, and wants global frontier models held to the same rules as Indian ones.

The government has signalled that new law is coming. IT Secretary S Krishnan said earlier this year that "the time is getting right" to look at AI regulation.

Where this lands for enterprises

I expect the hybrid view to win, and it's the one that lands hardest on enterprise AI teams. If binding rules arrive first for banking, healthcare and government, then the companies deploying AI in those sectors, not just the model builders, will carry the obligations: incident reporting, named accountability, evidence of testing before go-live.

Velamakanni made another point that's easy to skip past. He said any safety pact has to include "large deployers", meaning enterprises and public agencies, alongside model developers. That's where most AI in India actually runs.

What to do now doesn't depend on who wins the policy argument. Keep a register of the AI systems you run. Know who owns each one. Log incidents now, before a regulator asks for the history. You'll need all of that under a voluntary code or a binding one.

Source: Rediff / PTI: Why India Needs A Unique AI Safety Framework