Overview
- The NSA, which sources told Axios on Sunday had access to Anthropic’s restricted Mythos model, is using it despite the Pentagon’s “supply chain risk” label.
- Mythos is Anthropic’s most powerful AI to date and is described as strikingly good at computer security tasks such as finding exploitable software flaws.
- Organizations with access mostly point the model at their own networks to scan for weaknesses that hackers could use to break in.
- Access remains tightly limited to about 40 vetted groups with only a dozen named publicly, and the U.K.’s AI Security Institute says it also has the model.
- In February the Defense Department moved to cut off Anthropic and litigation followed, while a Friday White House meeting with CEO Dario Amodei was described as productive and focused on controlled government use outside the Pentagon.