Cisco has announced a major expansion of Cisco AI Defense platform at Cisco Live APJC 2026 in Melbourne, integrating Anthropic’s Claude Mythos into its adaptive red teaming framework to stress-test AI agents with frontier adversarial reasoning. The update, alongside new Splunk Agent Observability capabilities and Cisco IQ enhancements, is designed to give enterprises a continuous trust loop for securing, governing, and observing AI agents at machine speed.
Since Cisco AI Defense launched in January 2025, it has helped hundreds of customers identify critical flaws in AI models, applications, and agents before they reach production. The addition of Claude Mythos extends the platform’s red teaming capabilities with advanced reasoning that plans multi-step attack scenarios and translates results into targeted attack ideas for each system under test.
Cisco AI Defense Red Teaming With Claude Mythos
Claude Mythos brings realistic adversarial ideation to Cisco AI Defense, generating multi-step, contextual attack paths designed to surface subtle vulnerabilities across complex AI architectures. The integration operates through AI Defense’s structured interface, meaning customers have no open-ended prompt access to the underlying model. Every testing session is recorded, cryptographically tied to a named enterprise operator, and runs with time-bounded, revocable execution to ensure controlled engagement.
The framework is restricted to targets explicitly registered and authorized by the enterprise customer, preventing any scanning of arbitrary internet infrastructure. A dual-layer oversight model combines product-level controls within AI Defense with program-level safeguards and monitoring. Adaptive red teaming with Claude Mythos is coming soon for select enterprises.
Cisco is also introducing Tokenomics through Splunk Agent Observability, a cost governance capability that ties AI agent spend directly to output quality. Using data from Amazon Bedrock and Amazon CloudWatch, Tokenomics pairs each agent’s cost with evaluations that grade the quality and business impact of its work, giving organizations a consolidated view of AI return on investment. It will be available in private preview in December, with general availability in early 2027 via AWS Marketplace.
Alongside Tokenomics, Cisco is rolling out Eval-G1 Evaluation Models that deliver evaluation response times 2.5x faster than traditional large language models at up to 98% lower cost. These models are built by converting agentic traces and subject matter expert feedback into custom small language models capable of executing evaluations and guardrails in real time. A companion Evaluation Tuner allows teams to fine-tune evaluator models to enforce quality standards and steer agents at machine speed. Both the evaluation models and the Evaluation Tuner are generally available from this month.
Cisco IQ & Infrastructure Exposure Intelligence
Cisco is extending Cisco IQ, which serves close to 17,000 customers across more than 22 million devices, to connect vulnerability intelligence directly into enterprise security environments. A new Cisco Data Fabric integration ingests Cisco IQ vulnerability, lifecycle, and configuration data into the Splunk Platform via open APIs, enabling security operations teams and AI agents to correlate live threat detections with underlying network exposure inside Splunk Enterprise Security.
Model Context Protocol and ServiceNow connectors provide structured infrastructure context to supported AI agents and enterprise IT service management platforms, automatically translating exposure findings into asset-linked ServiceNow problem records. An AI-Powered Resilient Infrastructure Playbook surfaces critical lifecycle vulnerabilities, personalized alerts for known defects, and quantum-safe cryptographic readiness to help teams prioritize modernization roadmaps.
Pokdepinion: Cisco is using AI to hack AI so your AI doesn’t get hacked by someone else’s AI. AI-ception.
