US Officials and Banks Tackle Cyber Risks From Anthropic’s Mythos AI Model

Wall Street banks and top US financial officials are urgently addressing cybersecurity risks posed by Anthropic's powerful new AI model, Mythos, which can rapidly uncover and chain together hidden software vulnerabilities. US Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell summoned bank CEOs to Washington on April 7 for an emergency meeting, warning that Mythos signals a new era in cyber threats where attacks could unfold in minutes rather than months, according to Bloomberg reports.^[1] Meanwhile, banks have begun internal testing of the model at the urging of Trump administration officials, while Canada's major lenders convened with the Bank of Canada to discuss similar dangers.^[3]

Anthropic, a leading AI developer, has withheld Mythos from public release due to its advanced capabilities, restricting access to a select group of vetted organizations—including financial firms—with $100 million in usage credits provided for testing and patching critical systems.^[1][2] The model, described in a leaked Anthropic blog post as representing a "step change" in performance, excels at identifying multiple high-severity flaws in software and linking them into novel attacks, surpassing even top human cybersecurity experts.^[4] Developers have called it "terrifying" for its potential to exploit vulnerabilities in major operating systems and open-source infrastructure that underpins global digital systems.^[4]

This development has sparked widespread alarm across the finance sector, where secure transactions and data protection are paramount. Cybersecurity CEO Vlad Rikhter of Fencer warned on Bloomberg that companies must now defend at "machine speed," rethinking security amid AI-accelerated cyber warfare.^[4] In the US, the high-level DC summit highlighted fears of a faster-moving cyber arms race, with officials pushing banks to scan their systems proactively. Canadian financial leaders echoed these concerns in their Friday meeting, focusing on how Mythos could amplify threats to banking operations.^[3]

Not all reactions are unanimous, however. At the HumanX AI conference, some venture capitalists and founders favored Anthropic over rivals like OpenAI, viewing Mythos as a sign of Silicon Valley leadership.^[5] Others, including AI commentators cited by Business Insider, argue the risks are overhyped, suggesting Mythos offers no revolutionary leap beyond existing models and may reflect savvy PR rather than unprecedented danger.^[1] Security firm Aikido's analysis of 1,000 AI penetration tests found that while models like Mythos speed up attacks, they do not yet decisively favor attackers over defenders with deep system knowledge.^[5]

The stakes are high for the finance industry, which handles trillions in daily transactions and relies on vulnerable software stacks. A breach enabled by such AI could erode trust, trigger massive losses, and disrupt markets, affecting consumers, investors, and economies worldwide.^[2] Anthropic is donating $4 million to open-source security efforts, aiming to mitigate fallout by funding patches for flaws Mythos uncovers.^[1]

Looking ahead, selected participants will use Mythos to audit and fortify their infrastructure, including critical open-source projects.^[1] Experts like former Facebook security chief Alex Stamos describe two timelines: an optimistic one where known flaws are exhausted and fixed, or a pessimistic view of endlessly emerging vulnerabilities from superhuman AI reasoning.^[2] Financial regulators and firms must now balance innovation—such as using Mythos for defense—with the imperative to stay ahead of AI-driven threats, as discussions intensify in boardrooms and government halls.