Anthropic’s AI model Mythos identified vulnerabilities in highly sensitive US government computer systems during a classified testing exercise, according to multiple sources [1, 2, 3, 4]. The tests took place under a secret operation called Project Glasswing, carried out in cooperation with Washington intelligence agencies [1, 2, 3, 4].
In a June 11 congressional hearing, Senator Mark Warner revealed that NSA chief Joshua Rudd informed him Mythos broke into almost all classified systems within hours, not weeks. Warner said, “Mythos broke into almost all of our classified systems, not in weeks, but in hours” [1, 2, 3, 4]. However, an anonymous US official clarified that while Mythos found some vulnerabilities quickly, it could not necessarily exploit them at the same speed [1, 2, 3, 4].
Anthropic refused requests to allow US military use of its AI for domestic surveillance or fully autonomous weapons, emphasizing ethical limits on the model's deployment [1, 4]. Due to national security concerns, the US government placed Anthropic on a national security blacklist and ordered suspension of Mythos and another AI model, Fable’s, exports to all foreign nationals [1, 3, 4].
Ongoing disputes led the NSA to lose access to the Mythos AI model, restricting their ability to continue testing or mitigation efforts with it [1, 4]. Experts consider Mythos a potentially serious risk to public safety, national security, and critical global software infrastructure [2].
News outlets published reports detailing Mythos’s vulnerabilities test and related government disputes on June 24, 2026 [1, 2, 3, 4]. The US government’s export restrictions and Anthropic’s refusal to support certain military uses mark the latest developments in the controversy surrounding Mythos AI.
