Scientists from 34 Countries Want Frontier AI Air-Gapped and Are Demanding International Oversight
Around 400 scientists from 34 countries, including AI pioneers Geoffrey Hinton and Yoshua Bengio, published an open letter on Tuesday, October 13, 2026 demanding an international oversight body for advanced AI systems with cyber capabilities — complete with industry-wide containment standards, air-gapping of frontier models, and a registry of security incidents that would eventually be made public. The proposal comes a month after a voluntary safety accord at the White House, and has met immediate pushback from both the European Commission and Yann LeCun.
What the Letter Actually Proposes
According to Reuters' coverage (reporter Leo Marchandon, Gdansk), the scientists propose an international body with three main tasks: setting minimum containment standards that apply across the industry, maintaining a registry of significant security incidents, and providing independent oversight of the development of future frontier models.
The registry of security incidents would, according to the scientists, "eventually be made public," Reuters writes. The letter also highlights an immediate demand, independent of the longer-term body question: organizations developing or testing cyber-capable models should adopt high-assurance containment standards right away, while governments work toward a more formal oversight regime.
A central technical recommendation is air-gapping. The scientists believe frontier AI models should initially be kept isolated from external networks and placed under independent monitoring. How such air-gapping would work in practice, and how strict the minimum standards should be, is not specified in the available reporting.
Why Voluntary Pledges Aren't Enough, According to the Scientists
The core of the critique is that the industry's current safety regime rests on voluntarism. According to The News, the letter notes that top executives at companies such as OpenAI, Anthropic and Google have committed to voluntary safety measures — without any form of enforcement.
That points directly to an accord at the White House roughly a month earlier, where Anthropic, OpenAI, Google, Meta, xAI and Nvidia signed on to voluntary safety measures with no formal enforcement mechanism, according to Reuters. That is the gap the letter is meant to fill: a body that applies across companies and national borders, with standards that can actually be checked.
Part of a Wave of Governance Initiatives
The letter does not arrive in a vacuum. According to Cryptobriefing, it builds on a declaration from September 21, 2026, in which leaders from roughly 20–22 countries called for mandatory testing before deployment and proposed an international institution modeled on the International Atomic Energy Agency (IAEA). Among the participating countries were Finland, Norway, Canada, Australia, Germany and Singapore. Two names stood out by their absence: the United States and China did not take part.
The timeline is tight. On the Sunday before the letter was published, the Trump administration lifted its block on US labs giving Britain's AI Security Institute early access to models for safety testing, according to Reuters. The sources establish no causal link between that lifting and the letter — but together they show a governance landscape in motion, with voluntary national accords, multilateral declarations without the great powers, and now a scientist-driven demand for binding international oversight.
The Pushback: Brussels and LeCun
The proposal quickly met skepticism from two directions.
EU Commissioner Henna Virkkunen told Reuters on Friday that the union's AI Act is sufficient to handle the risks from misbehaving AI systems. The argument, in other words, is that binding regulation already exists — at the EU level — and that a new international body is unnecessary.
The heaviest symbolic resistance comes from Yann LeCun. He shared the 2018 Turing Award with precisely Hinton and Bengio, but did not sign the letter. According to Reuters, LeCun has argued that the recent security incidents stem primarily from engineering failures, rather than existential AI risks. That puts two of three Turing laureates on one side of the issue, and the third on the other.
It is worth noting that the disagreement concerns interpretation, not just solutions: the signatories see a risk that requires binding international structures; LeCun sees errors that can be corrected in development work; and Virkkunen sees a regulatory question that has already been answered in the EU. Neither Reuters nor the other sources attempt to judge between these positions — and the incidents the letter builds on are not publicly documented in the available coverage.
What Remains Unclear
Several questions remain open as of the current reporting.
No enforcement mechanism has been specified. The letter says what the body should do, but not how its standards would be enforced against states and companies that do not participate — a problem the September declaration already illustrated, with the US and China outside.
The letter's full text and complete signatory list have not been independently confirmed. Every detail — the count of around 400 scientists, the 34 countries, the named demands — comes via Reuters and secondary coverage, not from the letter itself.
The security incidents that allegedly motivate the proposal are referenced, but not identified or documented in the available reporting. It is the scientists' characterization of them that underlies the air-gapping demand and the incident registry.
And the political outcome is open: a body with real authority over frontier AI models would require support from precisely the countries that stayed outside the September declaration. For now, the letter is an initiative from scientists — not an institution.