Key facts
- Anthropic's Mythos AI model identified vulnerabilities in classified U.S. government computer systems within hours.
- The testing was a collaboration between Anthropic and U.S. intelligence agencies under Project Glasswing.
- The Trump administration restricted foreign nationals' access to Anthropic's Fable 5 and Mythos 5 models.
- Over 100 cybersecurity experts argued the directive could harm U.S. defenses.
An advanced artificial intelligence model developed by Anthropic, named Mythos, has identified vulnerabilities in highly sensitive and secure U.S. government computer systems during a testing exercise, a U.S. official told The Associated Press. The official, speaking anonymously, stated that the model found these flaws within hours.
The testing was conducted through Anthropic's Project Glasswing initiative, a collaboration aimed at securing critical software. Senator Mark Warner had previously mentioned the exercise, noting that the tool "broke into almost all of our classified systems, not in weeks but in hours," attributing this to NSA and U.S. Cyber Command head Gen. Joshua Rudd.
Despite this cooperation, tensions have risen between Anthropic and the Trump administration. The administration recently issued a directive requiring Anthropic to prevent foreign nationals from using its latest AI models, Fable 5 and Mythos 5, citing cybersecurity fears. Anthropic complied by disabling the models for all customers, though it believes the government's actions were unwarranted.
A group of over 100 cybersecurity executives from companies like Adobe and Nvidia has also urged the administration to lift the directive. They argued in a letter that while Anthropic's models are effective at finding software flaws, they are not uniquely capable, and restricting access could benefit U.S. adversaries.