Washington leaves open AI models untouched, and India stands to gain

The News

Washington has quietly drawn a line around the part of artificial intelligence it is least willing to police. According to reporting from Axios, carried by The Verge, the White House's new framework for assessing the cybersecurity risks of advanced AI systems relies on voluntary guidelines that pointedly exclude open models, the kind that anyone can download, inspect and run on their own hardware.

The document reportedly goes further than merely ignoring open systems. It states in plain terms that the framework cannot be used to restrict open models once they have been released. In other words, once the weights are public, Washington is signalling it will not attempt to claw them back.

The framework arrives after President Trump signed an executive order on AI, and lands at a moment when governments everywhere are struggling to supervise systems that improve faster than the rulebooks meant to contain them.

Why It Matters

The open-versus-closed question has become the central fault line in AI governance, and this framework plants a flag firmly on one side of it. The last time the United States set out a formal AI safety posture, the Biden executive order of October 2023, the instinct ran the other way, leaning towards mandatory reporting and red-teaming for the most powerful models. The pendulum has now swung to a deliberately hands-off stance for anything published openly.

For the industry, the signal is unambiguous. Developers who release open-weight systems, from Meta's Llama family to a long tail of smaller labs, gain regulatory breathing room in their largest market. Critics counter that a model you cannot recall is a model you cannot govern, and that excluding open systems from testing leaves a blind spot precisely where misuse is hardest to trace. Either way, the message to builders is that open weights carry lighter compliance baggage in the United States than closed, hosted alternatives.

Indian Angle

India's own AI ambitions lean unusually hard on exactly the category Washington has chosen to leave alone. The over 10,000 crore rupee IndiaAI Mission is built around shared compute and support for indigenous foundation models, and much of the country's home-grown work is explicitly open. Sarvam AI has put out open-weight models tuned for Indian languages, and Krutrim, the AI arm backed by Ola, has open-sourced Indic model work of its own. A permissive American posture towards open weights is a tailwind for these firms, both for the freely downloadable foundations they build upon and for any ambition to serve users abroad.

There is a regulatory lesson too. MeitY has already lived through its own version of this debate, issuing an advisory in early 2024 that unsettled startups before it was softened, a reminder that heavy-handed rules on under-tested models can chill exactly the ecosystem a country is trying to nurture. Yet India cannot simply copy the American shrug. With elections, deepfakes and financial fraud all colliding online, the harms that flow from ungoverned open models land squarely within the IT Rules and the emerging DPDP framework. The task for New Delhi is to keep the runway clear for Sarvam and Krutrim while still holding someone accountable when an open model is turned against ordinary users.

For Indian founders and investors, the takeaway is subtle but real: the regulatory cost of building on open weights just fell in the world's most influential AI market, sweetening the open-first bet much of Indian AI has placed.

FAQ

Does the framework restrict open-source AI models?

No. The reported guidelines deliberately exclude open models from the testing regime, and the document states it cannot be used to restrict them once they have been released. Closed, hosted systems are the framework's real focus.

Is the framework legally binding?

No. It sets out voluntary guidelines rather than mandatory rules. That means compliance rests on the goodwill of developers, which is one reason safety researchers have questioned how much practical force it will carry.

What does this mean for Indian model builders?

It is broadly favourable. Firms such as Sarvam AI and Krutrim that release open-weight, Indic-language models face a lighter regulatory burden in the United States, improving the economics of the open-first strategy that much of Indian AI has adopted.

Where can I read the original coverage?

The Verge reported the story, drawing on Axios's account of the framework. The full coverage is linked in the source attribution below.

This story was reported by The Verge. Read the full original coverage at The Verge.