Cisco announced on November 9, 2026, that it plans to add Anthropic’s Claude Mythos 5 to Cisco AI Defense’s adaptive red-teaming framework. The model is intended to help shape tests for specific AI applications, within engagements Cisco describes as authorized and tightly scoped. Cisco’s announcement sets out the planned integration and its testing controls.
Mythos is planned as part of a controlled, application-specific testing workflow—not as an open prompt interface for customers.
For context, NeoTeo’s earlier coverage of Anthropic’s Cyber Verification Program addresses Anthropic’s access tiers; Cisco’s announcement concerns a separate integration into AI Defense.
How adaptive red teaming differs from standard tests
Red teaming means testing a system by simulating attacks to look for weaknesses. Cisco describes standard tests as covering common AI risks, including prompt injection, jailbreaks, exposure of sensitive data and harmful-content generation. Its adaptive approach is designed to follow criteria written in plain language and tailor tests to a particular application.
| Testing approach | Focus | How tests are shaped | Claude Mythos 5’s planned role |
| Standard red teaming | Common AI risks, such as prompt injection and sensitive-data exposure | A broad set of tests addresses common risk categories | Not part of the announced integration |
| Adaptive red teaming | Risks and behaviors specific to an AI application | The framework analyzes the target, tests tailored attacks and adapts to its responses | Contribute reasoning, attack ideas and concrete payloads within the test process |
Cisco says the adaptive framework would analyze a target, plan tests, assess responses and report findings. Mythos would contribute reasoning and attack ideas to that process. In practical terms, the announced role is to help shape application-specific tests; the customer is not being offered an unrestricted interface for prompting Mythos directly.
The controls Cisco describes for an engagement
Cisco says each Mythos-enabled engagement would use a registered target the customer is authorized to test, a fixed job and objectives defined by a person. The activity would remain confined to that target, be attributable to a named operator and be recorded for audit. Those boundaries make the test’s target and purpose explicit while leaving the test process within AI Defense.
Cisco’s stated rollout plan
Cisco said on November 9 that Mythos-enabled testing was coming soon for select enterprises, with broader access planned for qualified customers. It gave no calendar launch date and said timing, scope and regional availability could change.