Anthropic’s Mythos AI model has reportedly uncovered vulnerabilities in classified US government systems.
As reported by the Associated Press, the vulnerabilities were identified in highly sensitive US government computer systems during a testing exercise. Anthropic, the company responsible for Mythos and its Claude family of AI models, had been working with the government’s intelligence agencies to conduct tests as part of Project Glasswing, a cybersecurity partnership that offered early access to the AI.
The source that spoke with the AP said that, although Mythos identified certain vulnerabilities within hours, it didn’t mean the model was able to exploit them within that time.
Democratic Senator Mark Warner of Virginia briefly mentioned the testing earlier this month before the Senate Committee on Banking, Housing and Urban Affairs.
“This tool broke into almost all of our classified systems, not in weeks but in hours,” Warner said, attributing the information to the head of the National Security Agency and US Cyber Command, Gen. Joshua Rudd.
The news comes as Anthropic and US government are in conflict, with the US government recently ordering the suspension of the AI company’s Fable 5 and Mythos 5 AI models for foreign nationals, citing national security reasons. It came just days after the models were first launched by Anthropic and after the Trump administration raised security concerns following their public release.
This was after Anthropic refused to allow the government to use its AI models for domestic surveillance and fully autonomous weapons systems.
Anthropic has acknowledged that its AI technology is extremely powerful but disagrees with the US government stepping in to block use of the model.
Related stories
The Government who cried Fable: How the US/Anthropic saga could be felt worldwide
Tesla reportedly developing ‘Megapod’ modular AI data centre hardware
