Twenty-Eight Countries, Including the U.S. and China, Sign the First Global Agreement on AI Risk

At a summit at Bletchley Park — where Britain broke Nazi codes — governments agreed that the most capable AI systems could cause 'serious, even catastrophic, harm' and committed to testing them together. It is a statement of intent, not a treaty.

Delegates from twenty-eight countries and the European Union signed a declaration this morning at Bletchley Park, the English country house where British codebreakers worked during the Second World War, agreeing that advanced AI poses risks serious enough to require international cooperation. The United States and China both signed, which is the point: on almost no other subject are the two currently agreeing to anything.

The document is short and careful. It says the most capable systems could cause "serious, even catastrophic, harm, either deliberate or unintentional," it names misuse and loss of control as concerns, and it commits the signers to identifying risks together and to supporting scientific research on safety. It creates no rules and no enforcement. Britain, which convened the summit, announced a new government institute to test AI systems before release, and the United States announced one of its own.

The two-day summit brings together an unusual mix — Elon Musk, the heads of OpenAI and Google DeepMind, the U.S. vice president, the Chinese vice minister of science — and the choice of venue is deliberate. Rishi Sunak, the British prime minister, wants his country to be where the world's AI safety conversation happens, and Bletchley is a reminder that this country has done consequential things with computers before.

Critics note that the summit is about the risks the AI companies themselves emphasize — future, catastrophic — and less about present ones, and that the companies were at the table while civil society groups were mostly outside it. Both things are true. It is also true that a year ago there was no table.