Cisco Puts Anthropic’s Claude Mythos Behind AI Defense Red Teaming

Cisco AI Defense adds Anthropic’s Claude Mythos to adaptive red teaming. Cisco’s SVP and GM of AI Software and Platform group DJ Sampath tells CRN why it’s ‘a game changer’ for partners securing AI agents.

Cisco AI Defense is one of the first AI security products to put Anthropic’s Claude Mythos behind its adaptive red teaming engine, the tech giant revealed on Monday at Cisco Live APJC in Melbourne.

AI Defense will now pair frontier-level attacker reasoning with safeguards, including testing only authorized targets, tying each engagement to a named operator, and full auditability, so red team exercises mirror real adversaries without endangering customer environments, DJ Sampath, senior vice president and general manager of Cisco’s AI Software and Platform group, told CRN.

“You need to really deeply understand each of [the] use cases so that you’re red teaming against those use cases to be able to discover those gaps,” said Sampath (pictured). “We’re moving from generic red teaming to adaptive red teaming, but more importantly, we’re using Mythos model to be able to do it.”

AI Defense, first introduced in 2025, is a tool that helps safeguard enterprises against the misuse of AI tools and data leakage, and protect AI models, applications and agents.

Cisco AI Defense offers standard and adaptive red teaming for AI agents and applications today. Standard tests cover thousands of common vulnerabilities, including prompt injection, jailbreaks, and sensitive data exposure. It can produce reports mapped to Cisco’s AI security framework and other security standards. Adaptive red teaming, however, goes further, Cisco said. Users can define custom criteria in plain language, and the framework plans and runs multi-step attacks tailored to their own applications. Claude Mythos adds attacker-level reasoning to that engine, with safeguards built for authorized testing, according to the San Jose, Calif.-based company.

Observability is critical to running AI agents safely, Cisco’s DJ Sampath said. To that end, Cisco is running Mythos in a sandbox environment with full control over what it can access, how it executes and how its sub-agents are spun up. That same visibility feeds “tokenomics,” which Sampath said has a security angle as well as a cost one.

“Attackers are going to use agents with tokens to attack their infrastructure,” he said. “That’s the only way you can defend an attack that is now being committed by an agent swarm. You’ve got to have a swarm yourself.”

To that end, AI Defense now supports Amazon Bedrock and Amazon CloudWatch so partners and users can see token spend and ROI in a single dashboard, Sampath said.

The Partner Opportunity

Cisco partners can use Mythos through AI Defense to red team every client environment that’s rolling out agents, Sampath said.

No other red teaming service that he knows of has a model that powerful, he added.

“We give access to Mythos via Cisco AI Defense, and we use that to help our clients secure their deployments of agents. It’s a game changer for [partners],” Sampath said.

AI Defense comprises the security capabilities within Cisco Cloud Control, the company’s unified operations platform for AgenticOps where human operators and AI agents manage IT infrastructure together. Cloud Control, introduced at Cisco Live in June, is the operating platform, while Cisco IQ is the support and success platform fed by that data, Sampath said.

Partners now have access to the Cisco IQ platform to identify customer risk, uncover refresh opportunities, and build managed services, which the tech giant announced last month at Partner Summit to the delight of partners in attendance.

AI Defense has approximately 400 to 500 customers since the offering was launched last year. Sampath said that “self-improving engines” are coming, as well as security for agents on laptops, in the cloud and at the edge.

“We’re expanding the capabilities of AI Defense every single week,” he said. “Whether your agents are running on your laptops, inside of the cloud, or on the edge, we’re going to find ways of being able to secure them,” Sampath said.

Adaptive red teaming with Claude Mythos is coming soon for select enterprises. Broader availability is planned for qualified customers, but dates, scope, and region availability remain subject to change, according to Cisco.