Cisco adds Anthropic's Claude Mythos to AI Defense for AI-agent red teaming

Cisco is adding Anthropic's Claude Mythos to its expanded AI Defense platform for red teaming AI applications and agents, it announced at Cisco Live APJC in Melbourne, according to its press release. Claude Mythos will generate contextual attack paths, while AI Defense will record testing actions and restrict tests to registered targets authorized by customers.

The integration will require authenticated operators and tie each test to a named user. Tests will run with defined objectives and limited sessions that operators can stop. Adaptive red teaming with Claude Mythos will become available soon to selected enterprises.

Cisco also introduced Splunk Agent Observability updates for tracking agent costs and output quality. Tokenomics will enter private preview in December and become generally available through AWS Marketplace in early 2027. Eval-G1 evaluation models and the Evaluation Tuner are generally available this month.

Updates to Cisco IQ will send vulnerability, product lifecycle and configuration data into the Splunk Platform. New connectors will provide infrastructure data to supported AI agents and create asset-linked ServiceNow records from exposure findings.