For several years now, Cisco has pioneered the AI security industry with a number of firsts. It started with the launch of AI Defense, the first enterprise AI security offering to combine asset visibility, supply chain scanning, algorithmic red teaming, and runtime guardrails in a single unified solution. When agents emerged and rapidly reframed the security conversation, we were the first to introduce adaptive red teaming and adaptive guardrail capabilities, enabling our customers to test for and protect against risks unique to their applications, organizations, and industries.
Today, we are extending that work. Cisco AI Defense is among the first AI security products to augment its adaptive AI red teaming framework with Anthropic’s Claude Mythos. We built an adaptive AI red teaming offering that pairs frontier attacker reasoning with responsible controls: authorization to test only targets the customer is permitted to test; attribution to a named operator; and full auditability. Red team engagements should reflect adversary capability without compromising the security and safety of our customer environments.
This comes at a consequential moment, as businesses are eager to adopt agentic technologies and security teams work to keep pace. Beyond accounting for model behavior, organizations must now account for how agents behave once connected to a complex, interconnected system of tools, skills, and sensitive resources. Every environment is different, and security teams need a way to find the risks specific to their own deployments before attackers do.
Since the launch of AI Defense, Cisco customers have leveraged its best-in-class red teaming capabilities to uncover critical safety and security vulnerabilities in their AI systems. By bringing Claude Mythos to AI Defense, we’re augmenting that system to deliver an advanced AI red teaming experience to customers.
Understanding Adaptive Red Teaming in AI Defense
To contextualize the impact of Claude Mythos, it’s important to first understand how Cisco AI Defense performs red team testing on agents and AI applications today.
AI Defense offers two types of red team testing: standard and adaptive.
Standard red teaming assesses AI systems for a wide range of common vulnerabilities. These are risks that virtually any agent or AI application would be susceptible to: prompt injection and jailbreaks, sensitive data exposure, and harmful content generation, for example. These standard tests cover thousands of different categories, content types, and adversarial techniques in minutes. Results are detailed in a comprehensive report that maps back to Cisco’s Integrated AI Security and Safety Framework along with standards from OWASP, MITRE, and NIST.
Standard red teaming cannot comprehensively test for the specific risks unique to a given organization’s application, which is why we developed and launched our industry-defining adaptive red teaming agentic framework in AI Defense. Users simply define custom testing criteria in plain language; our red teaming framework analyzes the target system, develops and tests attacks tailored to those criteria, and uses the target’s responses to refine its approach or change tactics. It then evaluates the results and furnishes a report of findings. AI Defense recommends follow-up actions while also enabling customers to protect against specific risks by building custom adaptive guardrails in Policy Studio.
Applying Claude Mythos 5 to Adaptive AI Red Teaming
Cisco customers exist across virtually every industry, and their uses of agentic AI are correspondingly diverse. Adaptive red teaming has equipped them to test for unique security vulnerabilities, as well as behavioral issues, brand risks, and other highly contextual concerns.
Bringing Claude Mythos to AI Defense augments the adaptive red teaming engine where the attacker’s reasoning matters most. As the engine works to understand a target system and plan tests, the model’s reasoning produces attack ideas. When those ideas become concrete payloads, the benefit is sharper still: the capabilities of Claude Mythos are offered under safeguards built for authorized testing.
The model sits inside AI Defense with a fixed job, a registered target, and objectives defined by the human user. While customers do not get an open interface to Claude Mythos, they get to inform the target and objective, and our agentic red teaming framework conducts the engagement within those bounds. We regard this as an early instance of a broader shift, in which the frontier capabilities most relevant to security are made available to defenders under verified, controlled access rather than withheld from everyone.
“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, Cybersecurity Lead at Anthropic. “By bringing Claude Mythos to Cisco AI Defense, Cisco can help security teams red team their agents with advanced reasoning that plans tests and generates attack ideas for each system.”
Cisco designed this deployment to operate against only the target a customer has registered and is authorized to test; it runs with a fixed job and human-defined objectives, has no open interface a customer could use to prompt it freely, and its activity is confined to the authorized target rather than the wider internet. Every action it takes is attributable to a named operator and recorded for audit. These product-level controls sit alongside the program-level safeguards and monitoring, so containment does not rest on any single layer.
What Comes Next
Standard and adaptive red teaming is already available in Cisco AI Defense today.
Adaptive red teaming with Claude Mythos is coming soon for select enterprises. Broader availability is planned for qualified customers on a when-and-if available basis. Dates, scope, region availability remain subject to change.
As agents become more capable, the tools used to test them must be equally capable. Adding Claude Mythos to AI Defense is an important early step toward keeping that true, and we will share what we learn as these engagements progress.
Learn more about AI red teaming and the other capabilities offered in Cisco AI Defense here.
