The Cybersecurity and Infrastructure Security Agency does not have access to Anthropic's powerful new Mythos Preview model, according to sources who spoke to Axios. This exclusion comes despite other government agencies, including the National Security Agency, reportedly using the tool. The AI model was designed to discover and exploit security vulnerabilities with unprecedented speed.
CISA is the federal agency tasked with defending critical infrastructure from digital threats. Its absence from the testing group raises concerns about its ability to understand emerging AI-powered attack methods. Industries the agency works with, including finance and energy, are reportedly worried about being overwhelmed by such threats.
Anthropic decided against a public release of Mythos due to its advanced capabilities. Instead, the company provided it to more than 40 companies and organizations for testing and defensive work. An Anthropic official stated the company had briefed CISA on the model's capabilities earlier this month.
The situation highlights a potential gap in the government's defensive posture as powerful AI tools proliferate. While some agencies are testing the technology, the primary cyber defense organization appears to be sidelined. This could complicate efforts to develop coordinated defenses and share threat intelligence across the public and private sectors.
A counterargument suggests that restricting access to such a dangerous tool, even for defensive purposes, could be a prudent security measure. Some experts might argue that the risks of the model itself being compromised or misused outweigh the defensive benefits of wider testing.