Anthropic’s Mythos Found Vulnerabilities in Classified U.S. Systems, Official Says
Summary
An anonymous U.S. official told the AP that Anthropic’s Mythos found vulnerabilities in sensitive government systems during testing, but said this did not show the model could exploit them. The testing and model restrictions drew industry debate.
Key points
- An anonymous U.S. official said Mythos identified vulnerabilities in highly sensitive government systems within hours during tests with U.S. intelligence agencies; the official said this did not establish that the model could exploit them.
- The official said testing took place through Anthropic’s Project Glasswing, an initiative intended to help secure critical software.
- Sen. Mark Warner separately said Mythos had broken into almost all classified systems, attributing the information to NSA and U.S. Cyber Command chief Gen. Joshua Rudd; the NSA declined to comment.
- The administration issued a directive requiring Anthropic to prevent foreign nationals from using its latest models. Anthropic said it disabled the models for all customers to comply.
- More than 100 cybersecurity experts and industry leaders urged the administration to lift the directive, arguing that Mythos is capable at finding software flaws and weaponizing exploits but is not uniquely capable.
Article Details
- Event Type
- Anthropic’s Mythos model was reported to have identified vulnerabilities in highly sensitive U.S. government computer systems during testing conducted with U.S. intelligence agencies through Project Glasswing.
- Impact
- The testing raised concerns about potential cybersecurity, national security, public safety, and economic risks from advanced AI models. The anonymous official said vulnerabilities were identified within hours but clarified that this did not mean Mythos exploited them. Sen. Mark Warner separately said the tool had broken into almost all classified systems, attributing that information to Gen. Joshua Rudd. Anthropic later said it disabled the models for all customers to comply with a government directive.
People
Donald TrumpSigned an executive order establishing a framework for the federal government to vet national security risks of advanced AI systems before public release.Joshua RuddIdentified as the head of the National Security Agency and U.S. Cyber Command, to whom Mark Warner attributed information about the testing. The article does not independently confirm Warner’s characterization.Mark WarnerThe Democratic senator described the testing at a June 11 Senate hearing, saying the tool had broken into almost all classified systems within hours; he attributed the information to Gen. Joshua Rudd.
Vendors
Products
Fable 5The administration issued a directive earlier this month requiring Anthropic to prevent foreign nationals from using its latest artificial intelligence models, known as Fable 5 and Mythos 5.Mythosintelligence agencies to conduct tests using the company’s Mythos model.Mythos 5The administration issued a directive earlier this month requiring Anthropic to prevent foreign nationals from using its latest artificial intelligence models, known as Fable 5 and Mythos 5.
Countries
Industries
Artificial intelligenceofficial told The Associated Press on Tuesday that one of Anthropic’s artificial intelligence models had identified vulnerabilities in highly sensitive and secure U.S.CybersecurityThat model is a limited version of the more advanced Mythos, to which the company has tightly limited access due to cybersecurity fears.Governmentgovernment computer systems during a testing exercise.