Bletchley Declaration: 28 countries and the EU agree on frontier AI risks

The Bletchley Declaration is a statement on frontier AI risks agreed at the UK's AI Safety Summit at Bletchley Park on 1 November 2023. The UK government said 28 countries, including the United States and China, and the European Union endorsed it.

What it is

The declaration defines frontier AI as highly capable general-purpose models, including foundation models, whose abilities match or exceed the most advanced systems of the day. Signatories committed to support an international network of scientific research on frontier AI safety and to meet again, with the Republic of Korea co-hosting a virtual follow-up and France hosting the next in-person summit. New Zealand joined on 23 October 2024.

What it requires

Signatories named cybersecurity and biotechnology as domains of particular concern, along with frontier systems amplifying disinformation. The text places a particularly strong responsibility for safety on developers of unusually powerful and potentially harmful frontier systems, including safety testing and evaluations, and asks relevant actors for transparency on how they measure, monitor and mitigate harmful capabilities.

What security leaders should do

Because signatories singled out cybersecurity as a frontier-risk domain, cyber-misuse testing is a fair topic to raise with any frontier model supplier. When choosing a frontier model provider, ask what cyber-capability evaluations it runs before release, what it discloses about the results, and how it restricts misuse of its models for intrusion or malware work. Record the answers as part of third-party risk review.