Cisco adds Anthropic's Claude Mythos to AI Defense to red-team customers' AI agents
Mythos will plan tests and generate attack payloads only against targets a customer has registered, with no open prompt interface; it is coming soon for select enterprises, with no date given.
Why it matters
Anthropic has kept Mythos away from general release because of its offensive cyber skills, so packaging it inside a vendor's red-teaming product is a new route to defenders. Cisco says the model gets a fixed job, a registered target and human-set objectives, and every action is tied to a named operator and logged. Cisco calls it an early instance of frontier security capabilities going to defenders under verified, controlled access. Watch for which customers get it first and what findings Cisco publishes.
Key facts
- Cisco blog published 9 Nov 2026
- Mythos added to AI Defense adaptive red teaming
- Runs only against customer-registered targets
- No open interface to prompt the model
- Every action attributed to a named operator and logged
- Coming soon for select enterprises; no date given