Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell convened an urgent meeting at Treasury headquarters on Thursday with the chief executives of the largest US banks to discuss Anthropic's Mythos model and its implications for financial sector cyber security. The CEOs of Citigroup, Morgan Stanley, Bank of America, Wells Fargo, and Goldman Sachs attended. JPMorgan's Jamie Dimon was the only major banking CEO absent. The meeting, reported by Bloomberg and CNBC, served two purposes simultaneously: warning the banks about the elevated threat environment that Mythos creates, and urging them to participate in the defensive testing the model enables.
Anthropic released Mythos in limited access last week, distributing it to 52 partner organisations through Project Glasswing rather than making it available publicly. The company described the model as capable of identifying and exploiting vulnerabilities across every major operating system and every major web browser. Limiting the release was explicitly a safety decision: Anthropic concluded that a model with that capability should not be publicly available before defenders had the opportunity to use it for patching. The government's response suggests that assessment was well-founded. Bessent and Powell's decision to personally convene the heads of America's largest financial institutions is not a routine precautionary measure. It signals that people at the highest level of the US financial regulatory apparatus regard Mythos as a significant change to the threat landscape.
The meeting was not purely defensive in tone. Alongside the warning, the US government urged banks to actively test their systems using Mythos, presumably through access arrangements similar to Project Glasswing. This is the defensive proposition that Anthropic has been making since the model's release: the same capability that makes Mythos dangerous in an attacker's hands makes it valuable in a defender's. A bank that uses Mythos to scan its own codebase can find and patch vulnerabilities before an adversary exploits them. The urgency from Bessent and Powell suggests the government believes that window is real and time-limited.
Reporting also revealed that Vice President Vance and Bessent had already questioned major technology companies about AI security risks before the Mythos release, indicating the government had advance notice of what Anthropic was preparing to disclose. That pre-release consultation, combined with Thursday's meeting, suggests a level of coordination between Anthropic and US government officials that goes beyond the typical relationship between a private AI company and its regulators. Anthropic appears to have briefed senior government figures on Mythos before the public announcement, and those figures then took steps to prepare the financial sector.
The market read on Mythos is also worth noting. JPMorgan analysts described the model as positive for the major commercial cybersecurity vendors, specifically the companies best positioned to help enterprises use tools like Mythos to find and fix their own vulnerabilities. The same capability that Bessent and Powell were warning about is, in the financial analyst's framing, a revenue catalyst for the security industry. That dual character, simultaneously a threat and a commercial opportunity, is characteristic of how the most capable AI tools are entering the economy.
The meeting raises questions that go beyond this particular model. If a single AI model release is sufficient to prompt the Treasury Secretary and the Fed Chair to personally convene the heads of the major US banks, what does that imply about the pace at which government institutions can track and respond to AI capability advances? Mythos is one model from one company. The models that come after it will be more capable. The process Bessent and Powell used this week, urgent meeting, manual outreach, individual CEOs in a room at Treasury, does not scale to a world where similar capability jumps happen quarterly.