Cisco adds Anthropic Claude Mythos to AI Defense for frontier red teaming and ships new Splunk Agent Observability (Eval-G1, Evaluation Tuner, Tokenomics)

At Cisco Live Melbourne (Nov 9-10), Cisco announced its AI Defense adaptive red-teaming framework will add Anthropic's Claude Mythos (coming soon for select enterprises) and shipped new Splunk Agent Observability capabilities: Tokenomics cost governance (private preview in December), Eval-G1 evaluation models, and the Evaluation Tuner. Red-team engagements run only against registered, authorized targets with named operator attribution and revocable, time-bound jobs.

Combines frontier adversary reasoning with measured agent ROI, giving security and ops teams a single trust loop to stress-test agents and tie token spend to output quality — a significant agent-governance capability from a top-tier vendor.

Security teams on or evaluating Cisco AI Defense / Splunk should assess Claude Mythos red teaming for pre-production agent testing; FinOps and ops teams should review Tokenomics and Eval-G1 for spend-plus-quality governance.