Anthropic Mythos Cyber Risk Draws Fed Attention
Anthropic Mythos' Claude Mythos Preview found zero-day flaws, prompting Fed/Treasury talks with banks and refocusing traders on bank cyber exposure.
KEY TAKEAWAYS
- Claude Mythos Preview can identify and exploit zero-day vulnerabilities in major operating systems and browsers.
- Fed and Treasury convened an urgent meeting with major bank CEOs to address Mythos cyber risk.
- Project Glasswing gives Mythos early access to roughly 40 trusted partners for coordinated fixes.
HIGH POTENTIAL TRADES SENT DIRECTLY TO YOUR INBOX
Add your email to receive our free daily newsletter. No spam, unsubscribe anytime.
Anthropic said in a blog post on April 10, 2026, that its Claude Mythos Preview model, part of Anthropic Mythos, can identify and exploit zero-day vulnerabilities. This disclosure prompted an urgent meeting this week between Federal Reserve Chair Jerome Powell, Treasury Secretary Scott Bessent, and major bank CEOs to discuss the cyber risks.
Mythos Capabilities and Project Glasswing
Anthropic announced Claude Mythos Preview and Project Glasswing in its April 10 blog post. Mythos Preview is a nonpublic general-purpose language model that can autonomously detect and exploit zero-day vulnerabilities across major operating systems and web browsers when directed by a user.
The model uncovered vulnerabilities including a 27-year-old bug in OpenBSD, subtle race conditions, kernel address-space layout randomization (KASLR) bypasses, and chained exploits capable of escaping sandboxes. It also identified logic flaws such as authentication bypasses and denial-of-service weaknesses in web applications.
Project Glasswing grants Mythos access to about 40 trusted partners, including Amazon, Apple, Google, and JP Morgan, enabling coordinated early fixes before public disclosure. According to Anthropic, over 99% of discovered vulnerabilities remain unpatched under its coordinated disclosure process, with fewer than 1% publicly detailed.
Anthropic cautioned that "the same improvements that make the model substantially more effective at patching vulnerabilities also make it substantially more effective at exploiting them." The company framed this as a short-term risk favoring attackers if frontier labs release models carelessly, but a long-term advantage for defenders through faster patching and hardened processes.
Fed and Treasury Meet Banks on Cyber Risks
Federal Reserve Chair Jerome Powell and Treasury Secretary Scott Bessent held an urgent meeting this week in Washington with major U.S. bank CEOs to address cyber risks linked to Mythos. Bank leaders and officials discussed how tools capable of both detecting and weaponizing flaws might alter patching timelines and operational exposure for large institutions.
JP Morgan CEO Jamie Dimon recently warned in a shareholder letter that AI will almost surely worsen cyber risk, highlighting tensions between the promise of faster defenses and the threat of new offensive capabilities.
Anthropic’s ongoing legal dispute with the Trump administration over AI use by the Department of Defense adds another layer of policy scrutiny as banks and regulators consider governance, access, and deployment choices for high-capability models. The company’s disclosures and federal engagement signal that financial firms and policymakers view this technology as an operational and oversight priority.
