U.S. advanced AI regulation leaves loopholes, allowing organizations such as Cisco to still access Anthropic’s strongest security model
The U.S. government’s export controls against Anthropic did not block access, with Dragos and Cisco confirming they can still access the Mythos preview. Mythos can identify thousands of vulnerabilities in mainstream systems and browsers during testing. To prevent misuse, Anthropic limited Mythos access to about 200 members of the Project Glasswing alliance in April; at the same time, they launched Fable 5, which removes cybersecurity task permissions. The U.S. requested the blocking of Fable 5 and Mythos 5 to prevent unauthorized foreign use, and Anthropic immediately shut down the public service of Fable 5 and paused pushing Mythos 5 to Glasswing members. The government did not mention Mythos preview in the directive, but Anthropic stated that the remaining models are unaffected.

