Anthropic's Mythos Finds Critical Flaws in Classified US Systems: Report
The exercise offers a glimpse into how frontier AI models are being deployed by intelligence agencies to identify and remediate security vulnerabilities.
News
- Anthropic's Mythos Finds Critical Flaws in Classified US Systems: Report
- Qatar Launches AI-Focused Scholarship Program to Build Future Digital Workforce
- China's LineShine Tops TOP500, Becomes World's Fastest Supercomputer
- Japan's Sakana AI Unveils Fugu, Claims Edge Over Claude Fable 5 in Coding
- Oracle Cuts 21,000 Jobs as AI Silently Reshapes Its Operating Model
- Nvidia’s New Cooling System Cuts Data Center Water Use to Near Zero—But Not AI’s
[Image: Chetan Jha/MITSMR Middle East]
Anthropic’s advanced AI model, Mythos, identified vulnerabilities in highly sensitive U.S. government computer systems during a recent testing exercise conducted with intelligence agencies, according to a report by the Associated Press (AP).
The exercise, carried out under Project Glasswing—a restricted governmeant initiative designed to identify and remediate critical software weaknesses before they can be exploited by adversaries—offers a glimpse into how frontier AI models are increasingly being deployed as cybersecurity tools.
The program gained public attention after Senator Mark Warner of Virginia referenced the tests during a congressional hearing this month. Citing information provided by National Security Agency Director Joshua Rudd, Warner said Mythos was able to “break into almost all of our classified systems” within hours rather than weeks.
However, according to the AP, a U.S. official clarified that the model’s rapid success reflected its ability to identify vulnerabilities, not necessarily exploit them.
The results highlight a growing reality for governments and enterprises alike: the same AI capabilities that accelerate software development can also dramatically accelerate vulnerability discovery. As organizations integrate AI into critical operations, cybersecurity strategies may increasingly depend on using AI systems to identify weaknesses before attackers do.
The disclosure comes amid a complicated relationship between Anthropic and the U.S. government. The company has resisted allowing its models to be used for domestic surveillance and fully autonomous weapons systems, creating tensions with national security agencies. Those tensions have intensified in recent weeks following government restrictions on the export of Anthropic’s latest Mythos and Fable models over national security concerns.
The reported suspension of NSA access to Mythos, first reported by The New York Times, highlights a broader policy dilemma of governments increasingly dependent on AI capabilities while simultaneously seeking tighter control. The challenge is to map how governments can govern technologies they increasingly rely on.
