AIster · The ChatGPT shock · Laws & rulings
The US and UK act on AI safety in the same week
The first coordinated attempt by governments to manage the risks of frontier AI
Key facts
- Date
- 30 Oct–2 Nov 2023
- Era
- The ChatGPT shock
- Interest
- Laws & rulings
- Threads
- Safety & governance
- US
- Executive Order 14110, signed 30 Oct 2023 (revoked 20 Jan 2025)
- UK summit
- AI Safety Summit, Bletchley Park, 1–2 Nov 2023
- Declaration
- Agreed by 28 countries, including the US and China, plus the EU
- Follow-up
- UK AI Safety Institute launched 2 Nov 2023
Within one week at the end of October 2023, the two governments most involved in frontier AI set out their first serious responses to it.
The US executive order
On 30 October President Joe Biden signed Executive Order 14110, “Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence.” Its most discussed section used the Defense Production Act to require companies developing powerful “dual-use foundation models” to report to the government on their training, on how they protect model weights, and on the results of red-team safety testing (deliberate attempts to make a model misbehave). Until formal definitions were set, the reporting applied to models trained using more than 10^26 operations (a 1 followed by 26 zeros). The order also directed the National Institute of Standards and Technology (NIST) to develop guidelines for red-teaming. It was revoked on 20 January 2025 by Executive Order 14148.
The Bletchley Park summit
On 1 and 2 November the UK hosted the first AI Safety Summit at Bletchley Park. Governments, AI companies and researchers attended. On the first day, 28 countries and the European Union agreed the Bletchley Declaration. Signatories included the United States and China, alongside countries such as Brazil, India, Japan, Kenya, Nigeria, Saudi Arabia and the United Arab Emirates.
The declaration said there was “potential for serious, even catastrophic, harm” from the most capable AI models, whether deliberate or unintentional, and singled out cybersecurity, biotechnology and disinformation. It said developers of unusually powerful systems had a particular responsibility for safety testing, and committed signatories to build a shared, evidence-based understanding of frontier AI risk. South Korea agreed to co-host a smaller virtual summit within six months, and France to host the next in-person summit a year later.
A new institute
On 2 November Prime Minister Rishi Sunak launched the UK AI Safety Institute, built from the government's Frontier AI Taskforce with Ian Hogarth as chair, to test new frontier models before and after release. The US, Singapore and Google DeepMind were among those agreeing to partner with it.
Why it matters
None of this was binding law. Even so, it was the first time so many governments, including both the US and China, jointly named frontier AI as a safety issue. It also set up a series of follow-up summits, starting with South Korea and France.
Questions people ask
What is the Bletchley Declaration?
A statement agreed on 1 November 2023 at the UK's AI Safety Summit, in which governments recognised the risk of serious, even catastrophic, harm from frontier AI and agreed to cooperate on understanding and managing it.
Which countries signed the Bletchley Declaration?
28 countries and the European Union, including the United States, China, the United Kingdom, France, Germany, India, Japan, Brazil, Kenya, Nigeria and Saudi Arabia.
Sources
- GOV.UK, Countries agree to safe and responsible development of frontier AI in landmark Bletchley Declaration, 2023-11-01
- Federal Register, Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence (Executive Order 14110), 2023-11-01
- GOV.UK, Prime Minister launches new AI Safety Institute, 2023-11-02