At its Melbourne Cisco Live event, the networking giant unveiled plans to incorporate Anthropic's Claude Mythos into AI Defense, a service designed to identify vulnerabilities in proprietary AI systems and applications. The integration focuses on red teaming—systematically attempting to breach AI defenses to uncover weaknesses before malicious actors exploit them. Initial availability will be limited to select enterprise customers, with no timeline announced for broader rollout.
The Mythos model will generate attack strategies and devise testing scenarios against registered AI applications and agents. Cisco researchers previously demonstrated its capabilities in July by using it to circumvent guardrails on 15 different chatbots. However, direct prompt access remains unavailable to customers; Mythos operates exclusively against systems that organizations have explicitly enrolled in testing. Each interaction is attributed to a specific individual, logged for compliance purposes, and can be terminated immediately. Cisco confirmed that Mythos cannot access or scan external internet content.
Enterprises are increasingly building AI agents into their core operations and want to test those agents thoroughly before putting them into production
Michael Moore, Anthropic's cybersecurity lead
The security industry is rapidly adopting Mythos as a testing tool. Palo Alto Networks introduced continuous testing capabilities using Mythos alongside OpenAI's cyber model in September. IBM announced Monday that it intends to integrate Mythos into its Concert Protect tool during early 2027.
Anthropic's strategy emphasizes embedding Mythos within partner platforms rather than distributing it as a standalone product. The company released Mythos 5.2 on Monday, maintaining its restriction to approved organizations only. This cautious approach reflects security lessons learned in April, when an unauthorized group gained access to Mythos through a contractor's credentials.
Alongside the Mythos integration, Cisco enhanced its Splunk suite with new capabilities for monitoring AI agent expenses and assessing output quality. A pricing analysis tool named Tokenomics will enter private preview in December.
Source: The Next Web
