Anthropic to brief global regulators on AI cyber risks in finance

Anthropic is preparing to brief the Financial Stability Board on a model that can find and exploit zero-day flaws in every major operating system and web browser, a capability that could let attackers move faster than banks and regulators can respond. The discussion will put one of Silicon Valley’s most watched AI companies in front of finance ministries and central banks represented at the watchdog, turning a frontier-model debate into a question of financial stability.

The briefing was requested by Andrew Bailey, the Bank of England governor who chairs the Financial Stability Board, as regulators wrestle with the possibility that advanced AI can uncover weak points in global finance faster than institutions can patch them. The board coordinates rules for G20 economies, giving the issue reach far beyond any single bank or market. Anthropic has not commented publicly on the planned discussion.

The risk centers on Claude Mythos Preview, which Anthropic’s April 7, 2026 system card described as its most capable frontier model to date, with a striking jump over Claude Opus 4.6. In an April red-team post, Anthropic said the model could identify and then exploit zero-day vulnerabilities when instructed to do so. The company said testing found thousands of high-severity vulnerabilities, a finding that helps explain why cyber specialists see the tool as both a defensive asset and a potential accelerant for attack campaigns.

Anthropic has already moved to frame the model as a security product as well as a risk. On April 7, 2026, it launched Project Glasswing, a defensive cybersecurity initiative built around Mythos Preview and backed by partners including Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA and Palo Alto Networks. Anthropic said it would commit up to $100 million in usage credits and $4 million in donations to open-source security organizations.

The company has also said it is in ongoing discussions with U.S. government officials about Mythos Preview’s offensive and defensive cyber capabilities. That follows earlier claims that Anthropic disrupted what it called the first reported AI-orchestrated cyber espionage campaign, and later identified a threat actor using Claude Code across reconnaissance, vulnerability discovery, exploitation, credential harvesting and exfiltration. Together, those episodes suggest the next AI fight is not just about better code or faster summaries. It is about whether a model can spot cracks in the financial system before the people responsible for defending it can close them.