Cisco plans to integrate Anthropic’s Claude Mythos artificial intelligence model into its AI Defense platform to help companies identify vulnerabilities in their AI applications and agents through simulated cyberattacks.
The company announced the integration at its Cisco Live event in Melbourne, Australia. The feature will initially be available to selected enterprise customers, although Cisco has yet to announce a rollout date.
Claude Mythos will be used to develop attack strategies and generate ideas for testing vulnerabilities in companies’ AI applications and agents, a cybersecurity practice known as red teaming.
In July, Cisco researchers used the model to bypass the safety restrictions of 15 chatbots, demonstrating its ability to identify weaknesses in AI systems.
However, customers will not have direct access to Mythos through open-ended prompts. Instead, the model will operate within predefined tasks and can only test systems that customers own and have registered for security assessments.
Cisco said every test would be logged and linked to an identified individual, providing accountability and oversight over the testing process.
Cisco said every test conducted using Mythos would be linked to an identified individual, recorded for auditing purposes and subject to termination at any time.
The model would also be restricted from scanning the wider internet, according to a company blog post.
“Enterprises are increasingly building AI agents into their core operations and want to test those agents thoroughly before putting them into production,” said Michael Moore, Anthropic’s cybersecurity lead.
Meanwhile, cybersecurity companies are increasingly incorporating Mythos into their security products.
Palo Alto Networks launched an always-on testing service using Mythos and OpenAI’s cybersecurity model in September, while IBM announced on Monday plans to integrate Mythos into its Concert Protect tool in early 2027.
The approach aligns with Anthropic’s strategy of making Mythos available through selected technology partners rather than offering customers direct access to the model.
Anthropic released Mythos 5.2 on Monday, maintaining access restrictions that limit its availability to vetted organisations.
