Cisco is enhancing its AI Defense capabilities by integrating Anthropic’s Claude Mythos model to conduct sophisticated red teaming attacks on customers’ AI applications and agents, aiming to identify vulnerabilities before malicious actors do.
- Claude Mythos facilitates planned attacks on registered AI systems only
- Every red team test is tracked to a specific individual for audit
- Integration launches first on select enterprise customers without open prompt access
What happened
Cisco announced at its Cisco Live event in Melbourne that it will integrate Anthropic’s Claude Mythos into its AI Defense solution. This addition aims to bolster the platform’s ability to conduct red teaming exercises, where simulated attacks are planned against customers’ AI applications and agents to uncover potential security weaknesses before actual attackers exploit them.
The Mythos integration will initially be available to select enterprise customers. Importantly, clients will not have open prompt access to the model. Instead, it will operate on a fixed scope targeting only AI systems that a customer owns and explicitly registers for testing. All testing activity is linked to a named person, with comprehensive logging and the capacity to halt tests at any time to maintain control and traceability.
Why it matters
As enterprises increasingly incorporate AI agents into critical operations, the need to rigorously test these systems for vulnerabilities has become urgent. By leveraging Claude Mythos within AI Defense, Cisco provides its customers with a methodical and accountable means to simulate adversarial attacks on AI models, helping to fortify their defenses before deployment.
Anthropic’s approach of embedding Mythos inside partner platforms rather than distributing it broadly aligns with security best practices, reducing risks of misuse. Furthermore, this collaboration highlights growing industry momentum: other leading vendors like Palo Alto Networks and IBM are also integrating Mythos into their cybersecurity products to enhance AI threat detection and response capabilities.
What to watch next
Cisco has yet to announce a general release date for the Mythos integration beyond initial availability to select enterprises. Observers should monitor how broadly Cisco extends access in 2027 and how effectively customers use these red teaming tools to address emerging AI vulnerabilities in production environments.
Additionally, Cisco is advancing complementary AI monitoring features, including an upcoming private preview for a tool named Tokenomics designed to track costs and assess performance quality of AI agents. These developments signal Cisco’s continued investment in a comprehensive AI security ecosystem addressing both risk mitigation and operational efficiency.