Mythos: The AI Model Redefining Cybersecurity Standards
Anthropic CEO Dario Amodei met with White House Chief of Staff Susie Wiles on Friday to discuss the company's new AI model, Mythos, as governments and financial institutions across three continents scramble to assess the cybersecurity risks posed by what the UK's AI Security Institute has called the most capable offensive cyber tool it has ever evaluated.
The meeting, described by both sides as productive, covered AI safety, cybersecurity, and U.S. AI leadership. It came at the end of a turbulent two weeks in which the model's capabilities have triggered emergency meetings among top financial regulators and banking executives in Washington, Ottawa, and London.
A New Threshold in Cyber Capability
The UK's AI Security Institute published an evaluation of Claude Mythos Preview on April 13 finding the model represents "a step up" over previous frontier models in cyber offense. In controlled tests, Mythos became the first AI system to complete a 32-step corporate network attack simulation from start to finish, succeeding in three out of ten attempts and averaging 22 of 32 steps — far outpacing any prior model. On expert-level capture-the-flag challenges, Mythos achieved a 73% success rate, tasks no model could complete before April 2025.
The UK government issued an open letter to businesses on April 15 warning that frontier AI cyber capabilities are now doubling every four months, double the pace recorded previously. "The window businesses have to get their defences in order is closing faster than anyone anticipated," the letter stated.
Financial System on Alert
Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell convened an emergency meeting with top U.S. bank CEOs at Treasury headquarters on April 7 to discuss risks Mythos poses to the financial system, according to Bloomberg and CNBC. Attendees included the CEOs of Citigroup, Morgan Stanley, Bank of America, Wells Fargo, and Goldman Sachs.
In Canada, the Bank of Canada and major lenders held their own meetings, with Governor Tiff Macklem warning that "Mythos has arrived, it's a lot more powerful than what came before. But something else will come that's even more powerful than that". Canadian Finance Minister François-Philippe Champagne told the BBC that Mythos dominated discussions at the IMF spring meetings in Washington.
Controlled Access Under Project Glasswing
Anthropic launched Project Glasswing on April 7, granting Mythos access to 12 founding partners — including Amazon Web Services, Apple, Google, Microsoft, JPMorgan Chase, CrowdStrike, Nvidia, and Palo Alto Networks — along with roughly 40 additional organizations that maintain critical software infrastructure. The company has committed up to $100 million in usage credits to support the effort and has said it does not plan to make the model generally available.
Security researcher Costin Raiu warned that legacy banking systems are especially vulnerable, telling Reuters that "a model like Mythos would have a field day finding exploits" in decades-old IBM-built banking infrastructure.
