Cisco and IBM will put Anthropic's restricted Claude Mythos model into their security products, Cisco to attack customers' own AI apps and IBM to hunt code flaws

Two large security vendors said on 9 November that they will build Anthropic's Claude Mythos, the version of its top model that goes only to vetted organisations, into their products.

Cisco is adding Mythos to AI Defense to red-team AI applications and agents. “Adaptive red teaming with Claude Mythos is coming soon for select enterprises,” its release says. Cisco says the testing runs only against targets the enterprise has registered, owns and authorised, cannot scan arbitrary internet infrastructure, and records each testing action for audit. “You cannot scale what you cannot secure, govern, or observe,” said Jeetu Patel, Cisco's President and Chief Product Officer. The release carries a Melbourne dateline of 10 November, for Cisco Live APJC.

IBM said it plans to bring Mythos's cyber reasoning into IBM Concert Protect, its exposure-management product. In IBM's plan Mythos finds unknown flaws in code, Concert judges which are exploitable, and IBM and Red Hat tools draft fixes for developers to approve. “General availability is planned for early 2027,” IBM says, adding that its plans may change.

Anthropic released Mythos 5.2 the same day, through its verification programmes. The two deals show how restricted cyber models are reaching companies: through security vendors that wrap them in their own controls.

• Confirmed Cisco's adaptive red teaming with Claude Mythos is coming soon for select enterprises and works only against targets they have registered, own and authorised. Cisco
• Claimed IBM plans to bring Mythos into IBM Concert Protect, with general availability planned for early 2027. IBM