A technology team at the U.S. Treasury Department is seeking to gain access to the Anthropic artificial intelligence model Mythos in order to find potential vulnerabilities in the banking industry. U.S. Treasury Chief Information Officer Sam Corcos is seeking to gain access to the model as soon as this week, according to people familiar with the matter. A previous report from the Financial Associated Press mentioned that Anthropic currently only provides this model to a few institutions.
People familiar with the matter added that Corcos has introduced the technology to the Treasury Department’s cybersecurity team and instructed the team to prepare in advance for potential threats that powerful AI systems may pose.
Last week, U.S. Treasury Secretary Bessant and Federal Reserve Chairman Powell urgently convened an emergency meeting with Wall Street executives to discuss a new round of cybersecurity risks that Mythos may cause.
Anthropic warned that the model could be used to launch cyberattacks if companies do not adequately test their systems and build defenses. At the meeting, regulators urged financial institutions to take Mythos seriously and use it to proactively detect system vulnerabilities.
If the news is true, it would reflect that Mythos has become so powerful that the U.S. government has had to elevate it to a strategic level - even though the U.S. Department of War (Defense Department) listed Anthropic as a "supply chain risk" earlier this year, the U.S. Treasury Department is still seeking to acquire its technology.
Corcos was a co-founder of the medical technology startup Levels and a member of Musk's "Government Efficiency Department." He was involved in pushing government agencies to reduce spending and was appointed chief information officer of the Treasury Department in mid-2025.
People familiar with the matter mentioned that before Anthropic was listed as a supply chain risk by the U.S. War Department, Corcos promoted the use of the company’s Claude AI tool within the Treasury Department’s technology team.
In testing Mythos, the Anthropic security team found that the model was able to identify and exploit vulnerabilities in "almost all major operating systems and major browsers" under user instructions. In one case, it even wrote a browser exploit that exploited four vulnerabilities in series.
Recently, Anthropic launched a test plan called "Project Glasswing" to allow a small number of institutions to participate in the trial of Mythos. The company said it is in continuous communication with all levels of government and stands ready to cooperate with regulatory agencies at the local, state and federal levels.
At the same time, many Wall Street banks have also begun testing Mythos internally. In addition to JPMorgan Chase, major financial institutions such as Goldman Sachs, Citigroup, Bank of America, and Morgan Stanley have also gained access to the model.