Terry KS MALAYSIA, 11 NOVEMBER 2026 – Cisco has unveiled major updates to its enterprise artificial intelligence portfolio, introducing Anthropic’s Claude Mythos model into Cisco AI Defense alongside new Splunk Agent Observability tools and automated Cisco IQ integrations. The synchronized updates are engineered to help organizations stress-test autonomous systems against frontier-grade adversarial tactics while tracking runtime performance, token expenditure, and underlying network exposure.
The expansion arrives as enterprises rapidly deploy autonomous agents connected to proprietary databases and external toolsets. According to recent global research from Splunk, 81 percent of IT operations and engineering professionals report heightened collaboration between observability and security departments to maintain control over these automated workflows.
Securing agents requires a continuous trust loop that balances machine-speed defensive controls with measurable business output, noted Jeetu Patel, Cisco President and Chief Product Officer. Patel underscored that expanding artificial intelligence implementations across core business processes is only viable when organizations can simultaneously secure, monitor, and govern every interacting model.
Originally launched in January 2025, Cisco AI Defense provides adaptive red teaming to reveal systemic weaknesses before applications reach production environments. Incorporating Claude Mythos adds advanced reasoning capabilities, allowing the system to construct realistic, multi-step attack simulations that model sophisticated threat actor behaviors across complex enterprise topologies.
To safeguard testing operations, Cisco has configured AI Defense with multi-tiered operational guardrails. The architecture restricts testing strictly to customer-registered and authorized assets, prohibits open-ended web scanning, cryptographically ties every action to a named operator, and maintains human-revocable runtime controls with detailed audit logging. Michael Moore, Cybersecurity Lead at Anthropic, highlighted that pairing Claude Mythos with Cisco’s defensive framework equips security personnel with the analytical reasoning needed to rigorously assess production-bound agents.
Beyond threat detection, Cisco is targeting operational efficiency through Splunk Agent Observability to help leadership evaluate return on investment. The upcoming Tokenomics feature, slated for deployment via AWS Marketplace, correlates compute expenditure across services like Amazon Bedrock and Amazon CloudWatch with task quality metrics, providing transparent cost-to-impact tracking.
The observability suite also introduces Eval-G1 evaluation models, which process runtime telemetry and expert feedback into specialized small language models. Operating up to 98 percent cheaper and two and a half times faster than standard large models, these lightweight evaluators enforce real-time quality standards across agent interactions. Accompanying the models is an Evaluation Tuner that allows operational teams to calibrate automated oversight against internal benchmarks.
To expedite threat remediation, Cisco is connecting Cisco IQ telemetry directly into enterprise workflows via the Cisco Data Fabric powered by Splunk. By processing posture and lifecycle data across more than 22 million managed devices into Splunk Enterprise Security, operations personnel can link active alerts to infrastructure vulnerabilities in real time.
New connectors for the open Model Context Protocol and ServiceNow automatically convert discovered exposures into structured service records, removing manual intervention from remediation paths. In addition, an automated infrastructure playbook alerts teams to software anomalies and quantum-safe cryptographic readiness. Liz Centoni, Cisco Executive Vice President and Chief Customer Experience Officer, noted that channeling dynamic vulnerability intelligence into centralized investigation workflows ensures defenders can transition immediately from threat awareness to coordinated risk mitigation.
The Eval-G1 models and Evaluation Tuner are generally available starting this month. Adaptive red teaming powered by Claude Mythos will launch shortly for select enterprise organizations, while the Tokenomics governance framework enters private preview in December ahead of broader marketplace release in early 2027.