US Treasury and Fed Sound Alarm for Bank CEOs Over Anthropic’s Controversial Mythos AI Model
US Treasury Secretary Scott Bessent and Federal Reserve Chairman Jerome Powell held an emergency meeting with top bank executives this week, issuing stark warnings about the cybersecurity threats posed by Anthropic’s newly unveiled artificial intelligence model. Two sources familiar with the high-level gathering confirmed the details of the closed-door session held in Washington.
The urgency of the meeting stems from Anthropic’s Mythos model, which the AI startup launched earlier this week but deliberately withheld from a wide public release. The company acknowledged it held back a broader rollout due to serious concerns that the model could expose previously undiscovered cybersecurity vulnerabilities across critical digital infrastructure.
What makes Mythos particularly alarming to regulators and security professionals is the sheer scale of its capabilities. Anthropic itself has stated that the model is capable of identifying and actively exploiting weaknesses found across every major operating system and every major web browser — a declaration that immediately sent shockwaves through the financial and technology sectors.
Emergency Briefings Reveal the Scale of Mythos’s Cyber Threat to Financial Institutions
Prior to this week’s high-profile gathering, Anthropic had already been in active dialogue with senior US government officials. Last week, the company disclosed it was engaged in ongoing discussions with Washington regarding Mythos’s “offensive and defensive cyber capabilities”, signaling that concerns about the model had been building behind the scenes for some time.
A third source close to the situation confirmed that Anthropic took a proactive approach, briefing senior government officials and key industry stakeholders well ahead of the model’s release. The company appeared to understand the weight of what it had built and moved quickly to bring the right people into the conversation before releasing it — even in a limited capacity.
The treasury-hosted meeting on Tuesday was specifically designed to ensure that America’s largest financial institutions are fully aware of the risks that Mythos and similar next-generation AI models could pose to their networks and customer data. According to one of the sources, the session was also intended to push banks to actively strengthen their defensive measures against potential AI-powered cyberattacks.
Timing played a role in how swiftly the meeting came together. Most CEOs of the nation’s biggest banks were already present in Washington for separate engagements, making it logistically straightforward to gather them on short notice for the urgent treasury briefing.
Bloomberg News reported that the chief executives of Citigroup, Morgan Stanley, Bank of America, Wells Fargo, and Goldman Sachs were all in attendance at the meeting, underscoring just how seriously the federal government is taking the threat posed by this emerging technology.
Anthropic has confirmed that access to Mythos will be strictly limited to approximately 40 technology companies for now. Industry giants Microsoft and Google are among those granted entry to the controlled rollout, suggesting that even in its restricted form, the model is being carefully managed to prevent misuse.
Goldman Sachs, Wells Fargo, and the Federal Reserve all declined to provide comment on the matter. The Treasury Department, the lending institutions involved, and Anthropic did not immediately respond to media requests for a statement.
The extraordinary nature of this government-led intervention signals a new chapter in the regulation and oversight of advanced AI models — one where federal agencies are no longer willing to wait for a crisis before acting. As AI capabilities continue to advance at a rapid pace, the pressure on both private companies and financial institutions to stay ahead of these evolving threats has never been greater.