Anthropic engineers are reportedly helping the NSA use and customize Mythos, an AI model built for cyber work. The Financial Times reported the NSA work. No public source shows Mythos running live attacks.
That distinction matters. Helping the NSA prepare a cyber model is sensitive enough on its own. Saying Anthropic is already helping run attacks would go beyond what the public record proves.
Financial Times NSA Mythos report
Why Mythos matters
Anthropic says Mythos-style models can cut the time and cost needed to find and exploit software flaws. That can help defenders find holes first. It can also make offensive cyber work easier if the same kind of model is used by an intelligence agency.
The company has warned that similar cyber models may soon be available from many AI companies. Its Project Glasswing expansion post says many rivals may have Mythos-class tools within 6 to 12 months.
Anthropic wants both roles
Anthropic sells itself as a safety-minded AI company, but it also wants national-security customers. Its public statement says it has worked with the US government and objected when the Department of War pushed for broader AI-use terms.
That is the uncomfortable trade. A cyber model powerful enough to help find serious software flaws is also powerful enough to matter to the NSA. Anthropic can say it wants safe use. The NSA use case still deserves close scrutiny.
