Cisco Adds Claude Mythos to AI Defense

Cisco Integrates Claude Mythos into AI Defense to Strengthen Red Teaming for Enterprise AI Agents

Cisco is adding Anthropic’s Claude Mythos to its AI Defense platform to improve red teaming, helping security teams simulate more advanced attacks against enterprise AI applications and agents before deployment. The update also expands observability for AI agents, giving organizations more visibility into how these systems behave in production-like environments.

What Cisco is adding to AI Defense

Cisco says Claude Mythos will augment AI Defense’s adaptive red teaming framework with stronger adversarial reasoning, allowing security teams to design more realistic, multi-step attack scenarios against registered AI systems. The goal is to uncover vulnerabilities in AI apps and agents that simpler tests might miss.

• Adaptive red teaming for AI applications and agents
• Multi-step attack path generation
• Testing focused on customer-owned and authorized targets
• Auditability through named operator attribution and logging

Why the Claude Mythos integration matters

AI agents are becoming more capable, but that also means the attack surface is growing. Cisco is positioning Claude Mythos as a way to mimic more creative, frontier-level attacker behavior, so enterprises can identify weak points earlier and reduce risk before systems go live.

The integration is designed with guardrails, including restricted testing against only approved targets, controlled execution, and full traceability for each engagement. That makes the feature useful for security teams that need both realistic offensive testing and strong governance.

How Cisco is handling security and control

Cisco emphasizes that Claude Mythos will not be freely promptable by customers inside AI Defense. Instead, it runs in a controlled environment with a fixed objective, limited access, and sessions that can be stopped on demand. Each engagement is tied to a named enterprise operator and recorded for audit purposes.

• Only customer-registered targets are tested
• Each action is tied to an authenticated operator
• Sessions are time-bound and revocable
• Testing is confined to the authorized environment

Observability for enterprise AI agents

Alongside red teaming, Cisco is also expanding agent observability, which helps teams inspect how AI agents use tools, spend tokens, and execute tasks. This is important because observability can reveal inefficiencies, risky behavior, and operational issues that are hard to catch with traditional monitoring.

By combining red teaming with observability, Cisco is targeting both sides of enterprise AI security: prevention before deployment and visibility after deployment.

What this means for enterprise security teams

For enterprises adopting AI agents, Cisco’s update suggests a shift toward more structured AI security operations. Security teams can use the new capabilities to test how agents respond under pressure, understand how they behave in real environments, and improve oversight across the AI lifecycle.

• Better pre-deployment validation of AI systems
• More realistic attack simulation for agents
• Improved governance and audit readiness
• Greater visibility into AI agent execution and usage

Conclusion

Cisco’s integration of Claude Mythos into AI Defense reflects a broader push to secure enterprise AI with more advanced testing and deeper observability. For organizations deploying AI agents, the update offers a clearer path to spotting vulnerabilities earlier, improving oversight, and building more trustworthy AI systems.