Anthropic to Brief FSB on Mythos After AI Solves Cyber Test in 3 of 10 Runs
Updated
Updated · The Guardian · May 18
Anthropic to Brief FSB on Mythos After AI Solves Cyber Test in 3 of 10 Runs
11 articles · Updated · The Guardian · May 18
Anthropic will brief the Financial Stability Board on Claude Mythos after the company withheld the model from public release over concerns its cyber capabilities could aid hackers.
3 of 10 attempts on AISI’s previously unsolved “cooling tower” test succeeded in the latest Mythos version, which the UK institute called a notable capability jump over the preview model.
Apple and JP Morgan are among the banks and tech groups given access to Mythos to probe weaknesses the model can uncover, rather than letting it circulate broadly.
The FSB’s interest comes as the IMF this month warned fast-moving AI developments are raising financial-stability risks and that uneven cross-border oversight could weaken the global system.
Could a single advanced AI model trigger the next global financial crisis?
Has the world's most powerful cyberattack AI already been stolen?
Can defensive AI win an arms race against models that find security flaws in minutes?
Anthropic’s Mythos Uncovers 1,000s of Zero-Day Flaws: AI-Driven Cyber Threats Force Global Financial Regulators to Act
Overview
Anthropic's advanced AI model, Mythos, has created an urgent cybersecurity challenge for the global financial sector by autonomously discovering thousands of hidden software vulnerabilities, including critical zero-day flaws. This unprecedented capability has alarmed financial authorities, prompting Anthropic to brief the Financial Stability Board on the risks posed by such frontier AI systems. Industry leaders recognize Mythos as a game changer, but its power also exposes systemic weaknesses that financial institutions struggle to address quickly. With Mythos operating under limited release, the financial world faces growing concerns about defending against rapidly evolving AI-driven cyber threats.