Britain Steps Up Defences Against AI Risks
The UK government is enhancing its efforts to safeguard against artificial intelligence (AI) risks, following a successful pilot programme. The initiative aims to identify vulnerabilities within government systems using frontier AI models.
Pilot Results and Findings:
A recent pilot project revealed 407 weaknesses across nine government organisations within a month, at a cost of £13,000 in tokens. This highlights the potential for AI to uncover hidden flaws in complex systems.
Expansion and Legislation:
The UK will expand this programme, leveraging advanced AI models to scan its own code for vulnerabilities. Additionally, the government is considering legislation on AI incident reporting and addressing loss-of-control risk.
Key Takeaways:
- Enhancing AI Security: The programme aims to make the UK one of the hardest countries to target using AI.
- Pilot Success: Two pilots have identified and remedied over a thousand potential vulnerabilities.
- Challenges: AI agents can produce findings faster than human review, and "finding isn't the same as fixing."
- Global Collaboration: While the White House has asked certain AI labs to withhold new models, the UK urges restraint on labs whose systems its institute cannot reliably assess.
Future Plans:
The next phase involves moving from public repositories to closed government systems, with AI playing a central role during the UK's G20 presidency in 2027.