OpenAI sets national security rules, and India is not on the list

The News

OpenAI has published a formal set of National Security Principles, laying out how it will allow governments to use its frontier models for defence and security work. The company presented the move as an effort to bring transparency to a fast-growing and unusually sensitive line of business.

The document lands as OpenAI expands its dealings with the United States government and what it describes as allied partners, with an explicit focus on two areas: cyber defence and biosecurity. In both fields, the company argues, advanced AI can meaningfully advantage defenders rather than attackers.

To lend the exercise outside credibility, OpenAI said it engaged David Kris, a veteran national security lawyer, to facilitate the process and provide independent judgement. It also ran internal listening sessions spanning its research, safety, policy and government-partnership teams before publishing.

The principles insist that increasingly capable systems must be deployed in ways that reinforce, in the company's words, democratic accountability, meaningful human judgment, and the rule of law. In practice, OpenAI's rules continue to prohibit deployments such as mass domestic surveillance, the direction of autonomous weapons, and fully automated high-stakes decisions.

Why It Matters

The significance is less in any single clause than in the direction of travel. Barely two years ago, frontier AI labs treated defence contracts as reputational poison. In 2018, Google walked away from the Pentagon's Project Maven after thousands of employees signed a protest letter, and it briefly published principles ruling out weapons work altogether. That taboo has now collapsed. OpenAI courting national security agencies in public, complete with a named legal advisor and a published rulebook, is the clearest sign yet that defence has become a mainstream commercial priority for the largest labs.

The framing also matters. By stressing cyber and biosecurity as defensive domains, OpenAI is trying to occupy the moral high ground while still opening the door to lucrative government spending. The unresolved tension is obvious: a principle is only as strong as its enforcement, and contractual promises about human judgement are notoriously hard to audit once a model is inside a classified environment.

Indian Angle

For Indian readers, the most telling detail is an absence. The allied partners OpenAI is leaning towards sit largely within the Western security architecture. India, despite the much-trumpeted iCET framework with Washington on critical and emerging technology, does not feature in this defensive inner circle. That gap should sharpen a debate already under way in New Delhi.

It strengthens the case for the government's sovereign-AI push. The IndiaAI Mission has funded compute and backed home-grown model builders such as Sarvam and Ola's Krutrim precisely so that sensitive workloads need not depend on a foreign lab's shifting terms of service. If national security access to the best models is negotiated country by country, India's insistence on indigenous capability looks less like protectionism and more like prudence.

There is a commercial reading too. Indian IT majors and defence integrators that hoped to resell frontier models into government tenders now face a supplier that reserves the right to police end-use tightly. MeitY, DRDO and the iDEX ecosystem will note that a model available for enterprise deployment may still be walled off for security missions, and plan procurement accordingly. Data localisation under the DPDP Act adds a further layer of friction for any cross-border defence deployment.

FAQ

What exactly did OpenAI announce?

A published set of National Security Principles describing how it will work with governments on defence and security uses of its models, alongside an expansion of partnerships with the United States and allied nations focused on cyber defence and biosecurity.

Is India an OpenAI national security partner?

Not on the evidence of this announcement. OpenAI points to the United States and allied partners without naming India, despite the iCET technology pact. Indian agencies would need separate arrangements, and the government's sovereign-AI programme suggests it is hedging its bets.

Does this mean OpenAI is building weapons?

No. The principles reaffirm bars on directing autonomous weapons and on mass surveillance, and emphasise defensive uses such as cyber and biosecurity. Critics counter that such promises are difficult to verify once models operate inside classified systems.

Where can I read the original announcement?

The full text sits on OpenAI's own website in the post titled Our approach to government and national security partnerships, linked below.

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