BY TYLER DURDEN
TUESDAY, APR 14, 2026 – 10:40 AM
The US Treasury Department’s technology team is actively seeking access to Anthropic PBC’s highly restricted Mythos AI model so it can begin hunting for software vulnerabilities, according to a person familiar with the situation cited by Bloomberg.
Treasury Chief Information Officer Sam Corcos briefed the department’s cybersecurity team on the technology last week and has directed efforts to gain access to the model "as soon as this week."
The request comes days after Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell summoned top Wall Street CEOs to an urgent meeting at Treasury headquarters. Executives were warned that Mythos and similar frontier AI models could usher in a new era of heightened cyber risk.Anthropic itself has cautioned that the model may be capable of powering sophisticated cyberattacks unless companies proactively test it against their own systems and build defenses ahead of any wider release.
At the meeting, bank leaders were strongly urged to take the model seriously and use it internally to detect vulnerabilities.
What Is Mythos and Why the Restrictions?
Anthropic introduced Mythos (also referred to as Claude Mythos Preview) as part of its new Project Glasswing initiative. In internal testing, the model demonstrated extraordinary offensive cybersecurity capabilities: it was able to identify and exploit vulnerabilities “in every major operating system and every major web browser when directed by a user to do so.” In one documented case, it wrote a web browser exploit that successfully chained together four separate vulnerabilities.
Project Glasswing brings together Amazon Web Services (AWS), Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks to address growing concerns within the cybersecurity community that AI models are now capable of discovering and exploiting vulnerabilities at a faster pace than humans can keep up with.
According to the post on Anthropic’s website, the model’s strong agentic coding and reasoning skills enable it to uncover and exploit security flaws when directed by the user that have existed for years, even decades without detection. Benchmarking results cited by the company suggest a notable performance gap between Mythos Preview and its previous models in cybersecurity-related tasks. –cxtoday.com
Source