Washington - U.S. Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell met with chief executives from several large American banks earlier this week to review cybersecurity risks associated with Anthropic's latest artificial intelligence model, Claude Mythos Preview. The conversation brought together executives who were already in the capital for a separate banking industry meeting.
The attendees included leaders from Bank of America, Citigroup, Goldman Sachs, Morgan Stanley and Wells Fargo. JPMorgan Chase's CEO was extended an invitation but was unable to participate.
The meeting focused on concerns that Anthropic's Claude Mythos Preview model demonstrates an enhanced capacity to identify cybersecurity weaknesses that could potentially be exploited by malicious actors. Company communications noted that the model was distributed to a narrowly selected set of partners this week, including several large technology firms, with the stated aim of giving those partners an early opportunity to address any vulnerabilities.
According to the information released by the company, this represented the first instance in which Anthropic limited the initial availability of a model at launch. Anthropic described Mythos as a general purpose model whose capabilities are not confined to cybersecurity tasks.
Context and participants
Officials from the Treasury and the Federal Reserve engaged directly with bank leadership to surface and discuss the security implications tied to the model's capabilities. The presence of top executives from major banks underscored the priority placed on understanding potential operational and risk exposures.
Distribution and purpose
Anthropic provided Claude Mythos Preview to a selected group of partners this week, naming several large technology companies among them. The stated purpose of that limited release was to allow partners to receive early notice of any vulnerabilities and take steps to secure systems.
Model characterization
Anthropic characterized Mythos as a broadly capable model with uses beyond cybersecurity. The precise scope of those additional capabilities was not further detailed in the information available.
Summary takeaways
- Regulators and bank executives met to assess cybersecurity risks connected to a new AI model.
- The model was shared on a limited basis with select partners to enable early mitigation of vulnerabilities.
- Anthropic labeled the model as general purpose with capabilities beyond cybersecurity, but did not provide additional operational specifics.