For one day, America
and China agreed
about AI.

A statement on artificial intelligence safety agreed by 28 countries and the European Union at Bletchley Park in the United Kingdom on 1 November 2023, during the AI Safety Summit. It runs to roughly 1,300 words and sets out a shared understanding of AI's opportunities and risks, a definition of “frontier AI”, and a two-part agenda for international cooperation.

It is the first time a large group of governments — including both the United States and China — agreed a common text on AI risk. It remains the high-water mark of international AI consensus.

Twenty-eight countries and the European Union, listed alphabetically in the Declaration itself: Australia, Brazil, Canada, Chile, China, the European Union, France, Germany, India, Indonesia, Ireland, Israel, Italy, Japan, Kenya, the Kingdom of Saudi Arabia, the Netherlands, Nigeria, the Philippines, the Republic of Korea, Rwanda, Singapore, Spain, Switzerland, Türkiye, Ukraine, the United Arab Emirates, the United Kingdom, and the United States of America.

New Zealand joined the commitment on 23 October 2024, almost a year later, taking the total to 29 countries and the EU.

Yes — and that is the single most consequential fact about it. China signed alongside the United States and the European Union, and every country represented at the summit endorsed the text. The list is alphabetical, so China appears immediately before the European Union, which appears immediately before France.

It has not been repeated. At the Paris AI Action Summit in February 2025 the United States and the United Kingdom declined to sign the final statement — the two countries that convened and championed the Bletchley process.

No. It is a declaration — a statement of shared intent — not a treaty. It creates no obligations enforceable against any signatory, establishes no dispute mechanism, imposes no penalties and contains no compliance test. Its verbs give it away: affirm, recognise, resolve, encourage. There is not a single “shall” in it.

Nobody has ever breached the Bletchley Declaration, because it is not possible to. Its significance was diplomatic rather than legal.

Six things. A definition of frontier AI. A shared statement of the risks its signatories were most concerned by — misuse and loss of control, especially in cybersecurity and biotechnology, and the amplification of disinformation. A particular responsibility on the actors developing the most capable systems. A two-part agenda: identifying risks of shared concern through shared scientific understanding, and building respective risk-based national policies. Support for an internationally inclusive network of scientific research on frontier AI safety. And a commitment to meet again in 2024.

As highly capable general-purpose AI models, including foundation models, that could perform a wide variety of tasks — and, importantly, also relevant specific narrow AI that could exhibit capabilities that cause harm — which match or exceed the capabilities present in the most advanced models of the day.

Two things make that definition durable. It is capability-based rather than technology-based, so it does not expire when the architecture changes. And it expressly reaches narrow AI, which almost every summary of it omits.

At Bletchley Park in Buckinghamshire, England, on 1 November 2023, during the AI Safety Summit held on 1–2 November. The venue was the point: Bletchley Park was the home of British codebreaking during the Second World War and the birthplace of modern computing, and the choice was a deliberate piece of symbolism about international cooperation on a transformative technology.

The signatories kept their promise to meet again. The Republic of Korea co-hosted a second summit in Seoul in May 2024, producing the Seoul Declaration and the Frontier AI Safety Commitments, under which major AI developers published frontier safety frameworks. A network of AI Safety Institutes began to form.

France hosted the third in February 2025 — renamed the AI Action Summit, with “safety” dropped from the title. The United States and the United Kingdom declined to sign its final statement. The Bletchley consensus lasted roughly fifteen months.

Yes. It is © Crown copyright and published under the Open Government Licence v3.0, which permits anyone to copy, publish, distribute and adapt it, including commercially, provided the source is acknowledged and the licence identified. That is unusually permissive for a document of this importance, and it is why the Declaration is quoted so widely and so accurately.

The authoritative version is on GOV.UK, published by the Prime Minister's Office, the Foreign, Commonwealth and Development Office and the Department for Science, Innovation and Technology.

None, directly. It imposes nothing on companies. What it did was set the vocabulary and the direction that the binding instruments then followed — the EU AI Act's treatment of general-purpose models, the AI Safety Institutes and their evaluations, and the ISO/IEC standards that turn “trustworthy and responsible” into something auditable.

So the honest answer is that the Declaration will never be enforced against you, and the things it started very much will be. The EU AI Act › · ISO/IEC 42001 ›

Because it is the proof that it was once possible. Twenty-nine parties who agree on almost nothing — the United States and China, Israel and Saudi Arabia, Ukraine and the EU — agreed in public, with their names attached, that frontier AI poses risks serious enough to warrant coordinated attention.

Nothing since has matched it, and the vocabulary it fixed is still the vocabulary everyone uses. Whether the moment returns is one of the more consequential open questions in technology policy, and the Declaration is the benchmark against which any future attempt will be measured.