Anthropic assists US security agency’s use of Mythos for cyber attacks

Anthropic is helping the US National Security Agency deploy its powerful Mythos AI model for offensive cyber operations, embedding engineers inside the agency despite an ongoing legal battle with the Pentagon.  

The San Francisco-based company had installed about half a dozen staff within the NSA as so-called forward-deployed engineers to guide the use of the technology and customise models for specific applications, two people familiar with the arrangement said.

It remains unclear whether Anthropic’s engineers are assisting the NSA in active operations. However, one person close to the situation said Mythos would be useful for infiltrating the networks of nations such as China or Iran.

“The best way to build a good defence is to build a good attack,” said a person close to Anthropic, who argued that adversaries are probably building their own AI-driven offensive technology. “If [Mythos] is not used to build attack agents, adversaries will find a way to do it.”

The arrangement comes despite the Silicon Valley start-up’s legal battle with the defence department, which includes the NSA, over how its technology is used in warfighting.

Anthropic tried to restrict the US government’s use of its Claude AI models for mass surveillance of American citizens and lethal autonomous drones. That led the Pentagon to label the AI lab a “supply-chain risk”, a first for a US business. Anthropic has sued over the designation, which could force it to end contracts with organisations that work with the US military.

Since the dispute erupted, Anthropic unveiled Claude Mythos, triggering concern among governments, financial institutions and IT companies over its ability to detect and exploit software vulnerabilities.

Earlier this week, Anthropic announced it would distribute Mythos to 150 organisations across 15 countries, vastly expanding access beyond the US and UK. It was initially made available only to a select number of US-based organisations when it launched in April.

The move comes after Anthropic filed for an initial public offering that could value the company at more than $1tn, underlining the growing commercial and geopolitical significance of AI to national security.

Since Mythos’ release, Anthropic rival OpenAI has released a model with similar capabilities. Experts have warned that advanced AI models enable hackers to exploit computer systems but can also serve as a powerful tool against adversaries.

Anthropic and the US defence department both declined to comment.

The company has worked closely with the US government on the wider rollout, while the Trump administration has embedded the model across agencies.

In February, the FT reported that the Pentagon was seeking to create AI-powered cyber tools to identify infrastructure targets in China as part of an effort to improve US capabilities in any future military conflict with Beijing.

On Tuesday, President Donald Trump signed an executive order outlining a voluntary framework in which AI companies can submit their new models for security reviews before they are publicly released.

The order directs federal agencies to develop methods to evaluate the cyber capabilities of AI systems and to establish an “AI cyber security clearinghouse” to share information and strengthen defences.