AI security officials test Anthropic cyber threat as Bank of England to convene chiefs
AI minister Kanishka Narayan has said that Anthropic’s new cyber model was the most powerful seen yet as finance chiefs from the Bank of England and Treasury are set to meet next week to discuss the threats posed to the financial sector.
Narayan revealed that Claude Mythos, a new model that was held back from release due to its danger in finding unknown bugs and potentially allowing cyber attacks to be organised, has been tested by officials at the AI Security Institute (AISI).
He said that researchers at the state-backed body found it to be the “most capable model we’ve ever evaluated for cyber” and that it was the first model to complete its cyber-range in full, allowing it to go through systems and identify vulnerabilities.
Narayan added: “We’ve taken action based on our findings.”
He later added that it completed a “cyber range” that simulated a “corporate network” rather than a wider range.
Bank of England to lead meeting on AI model
The tool has added to fears across countries in the West that cyber systems are more vulnerable to attacks from AI models than ever before.
Bank chiefs from the likes of Citigroup, Goldman Sachs and Morgan Stanley in the US were summoned by treasury secretary Scott Bessent earlier this week.
It has been reported that bosses at the Bank of England, Treasury, Financial Conduct Authority and the National Cyber Security Centre will attend a meeting next week to discuss Mythos.
The Bank’s risk chief Duncan Mackinnon will chair the gathering of the Cross Market Operational Resilience Group, according to The Telegraph.
Anthropic bosses gave AISI early access to stress-test the model as well as multiple key tech companies, including Google, Nvidia, Apple and Amazon.
The AISI was founded under Rishi Sunak as part of an effort to boost AI research within the public sector.
The Department for Science, Innovation and Technology was given a £2bn increase in funding by Chancellor Rachel Reeves in the years until 2030.
The cash injection will fund research and encourage startups to build homegrown AI capabilities.
Government officials also hope they can persuade Anthropic’s chiefs to announce an expansion in the UK after the company refused to allow its technology to be used for fully autonomous weapons or mass.
The Trump administration attempted to designate the company a “supply chain risk” but a federal judge granted an injunction to prevent the ban on Anthropic.
London mayor Sadiq Khan has written to Anthropic’s chief executive Dario Amodei asking him to consider a move to London.