US Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell called the heads of America’s biggest banks to an urgent, short-notice meeting at the Treasury Department on Tuesday. The agenda wasn’t interest rates or tariffs—it was Anthropic’s newest AI model and the cybersecurity risks it poses to the financial system.The CEOs of Bank of America, Citigroup, Goldman Sachs, Morgan Stanley, and Wells Fargo were all in attendance. JPMorgan Chase CEO Jamie Dimon was invited but couldn’t make it. Every bank summoned is classified as systemically important by US regulators. The meeting was first reported by Bloomberg.At the centre of the discussion was Claude Mythos Preview—a new, unreleased AI model from Anthropic that can find and exploit software vulnerabilities better than nearly any human.Anthropic says the model has already uncovered thousands of severe, previously unknown flaws across every major operating system and web browser. One vulnerability in OpenBSD—widely regarded as one of the most secure operating systems in existence—had gone undetected for 27 years. Another, in the widely used video tool FFmpeg, sat in a line of code that automated testing tools had hit five million times without catching the problem.
Anthropic’s Project Glasswing wants to fix the flaws before bad actors find them
On Wednesday, Anthropic announced Project Glasswing—a defensive cybersecurity partnership with Amazon, Apple, Google, Microsoft, JPMorgan Chase, and others. The goal is to let these firms use Mythos to find and patch vulnerabilities before similar AI capabilities spread to attackers.Anthropic is putting up to $100 million in usage credits behind the effort and has donated $4 million to open-source security organisations. It has no plans to release Mythos to the public.
A $2 trillion stock wipeout, and now Wall Street gets summoned
Mythos isn’t the first time Anthropic has rattled markets this year. The company’s earlier releases—including Claude Opus and its agent-building tools—helped trigger a roughly $2 trillion selloff in enterprise software stocks, as investors confronted AI’s potential to gut the per-seat SaaS licensing model. If 10 AI agents can do the work of 100 employees, you don’t need 100 software subscriptions. The rout has been dubbed the “SaaSpocalypse.”When details of Mythos then leaked in late March through a configuration error at Anthropic, cybersecurity stocks took a further hit as the market processed a new fear—that AI could also commoditise the very security tools meant to protect against it.In his annual shareholder letter, Dimon wrote that cybersecurity remains one of JPMorgan’s biggest risks and that AI will likely make it worse, requiring significant investment in defence. Anthropic says it has been in ongoing discussions with US government officials about Mythos’s offensive and defensive capabilities. The Treasury Department, the Fed, and all the banks involved declined to comment.
