In a move that underscores the growing intersection of AI and financial stability, Anthropic has agreed to brief the Financial Stability Board (FSB) on the cyber‑security risks posed by its Claude Mythos model. The announcement comes after the model demonstrated unprecedented autonomous hacking capabilities, completing a previously unsolved cybersecurity test—known as “cooling tower”—in three out of ten attempts. This marks a significant escalation in frontier AI’s ability to perform complex cyber tasks, with the Autonomous Systems Institute (AISI) noting that the length of cyber tasks these models can complete autonomously has doubled in mere months, not years.
The FSB, which includes senior officials from central banks and finance ministries across major economies, will assess the implications of these findings for global financial stability. Anthropic has not released the Mythos model publicly, citing concerns over misuse, but has shared it with select tech firms and banks, including Apple and JPMorgan. The briefing reflects growing regulatory attention to AI’s dual-use potential—where models designed for productivity can also be weaponized for cyber threats.
This development follows broader warnings from the International Monetary Fund (IMF) about rising financial stability risks tied to rapid AI advances. As AI systems become more capable and autonomous, regulators and financial institutions are grappling with how to balance innovation with systemic resilience. Anthropic’s proactive engagement with the FSB may set a precedent for responsible disclosure and oversight in the AI‑finance nexus.