Anthropic agrees to update FSB on vulnerabilities identified by Mythos

Anthropic has agreed to brief members of the Financial Stability Board (FSB) on critical cyber vulnerabilities discovered by its Claude Mythos Preview AI model. The decision follows a request from Bank of England Governor and FSB Chair Andrew Bailey. The FSB comprises central bankers, finance ministry officials, and securities regulators representing G20 countries, including the UK, US, and China. 

There is growing concern among FSB members about the systemic risks that advanced AI models pose to the global banking infrastructure. Anthropic has faced increasing demands for access to its model, but access currently remains tightly restricted to around 40 organisations. Following a request from the White House, Anthropic has agreed to pause broader distribution. 

In related news, the Financial Conduct Authority (FCA), the Bank of England (BoE) and HM Treasury issued a joint statement on cyber resilience and the threats posed by frontier AI models. The statement warns that the capabilities of models being used maliciously could "amplify cyber threats to firms’ safety and soundness, customers, market integrity, and financial stability." 

£ - This Financial Times article requires a subscription.

A version of this article is available without subscription in The Guardian.

Training Announcement: The BCS Foundation Certificate in AI examines the challenges and risks associated with AI projects, such as those related to privacy, transparency and potential biases in algorithms that could lead to unintended consequences. Explore the role of data, effective risk management strategies, compliance requirements, and ongoing governance of the AI lifecycle and become a certified AI Governance professional. Find out more.