Cisco and IBM build Anthropic's Claude Mythos into their security products

AI Defense already runs adaptive red teaming, where security teams build multi-step attacks against their own AI models, applications and agents before they reach production. Mythos will plan the tests and turn the results into attack ideas for each system[1].

The access is tightly fenced. Mythos only works against targets the customer has registered and is authorised to test, cannot scan arbitrary internet infrastructure, and every engagement is authenticated, tied to a named operator, time-bound and recorded for audit. Customers get no open prompt access to the model[1]. Cisco calls this an early example of frontier security capabilities being given to defenders under verified, controlled access rather than withheld from everyone[2].

"Enterprises are increasingly building AI agents into their core operations and want to test those agents thoroughly before putting them into production," said Michael Moore, cybersecurity lead at Anthropic[1].

Adaptive red teaming with Mythos is coming soon for select enterprises, with broader availability planned for qualified customers[2]. The Register reports it will be offered under existing licences at no extra cost[4]. Cisco also announced Splunk Tokenomics, which ties AI agent spending to the quality of their work, and new Eval-G1 evaluation models that it says run up to 98% cheaper than a commercial LLM[1].

IBM's plan points Mythos at application code. The model would reason across customer-selected repositories to find previously unknown weaknesses, including logic flaws and potential zero-days with no existing signature or CVE[3]. Findings would flow into Concert Protect alongside traditional scanner results, where its Proof of Exploitability feature checks whether attackers could actually use them, and Concert Protect with Red Hat Lightwell generates fixes that developers review and approve through pull requests[3]. General availability is planned for early 2027[3].

Anthropic has said there is a long queue of organisations wanting Mythos, and routing it through Cisco and IBM lets far more defenders use it without the model itself being handed out[4]. That is the pattern to watch: the most dangerous AI capabilities reaching customers as fenced features inside products they already buy, rather than as open model access.