Anthropic's Bug-Finding AI Mythos Integrated into Cisco and IBM Security Products
Anthropic's advanced AI model for identifying software vulnerabilities, Mythos, is now being incorporated into security offerings from industry giants Cisco and IBM.
Anthropic is expanding the reach of its powerful bug-finding AI, Mythos, by enabling major cybersecurity players Cisco and IBM to integrate its capabilities into their respective product lines. This move signifies a growing trend of leveraging cutting-edge AI for proactive security testing and vulnerability discovery.
Cisco announced its integration at a recent event, detailing how Mythos will super-charge the red teaming tools within its AI Defense suite. This suite is designed to identify weaknesses, inappropriate activity, and potential threats within AI models, agents, and the applications they power. By incorporating Mythos, Cisco aims to provide its customers with a more robust method for testing their AI setups and uncovering vulnerabilities that might otherwise be missed by traditional security measures.
Jeff Schultz, Cisco's senior vice president for portfolio strategy and marketing, highlighted that AI Defense already employs AI-powered red teaming, but Mythos will serve as an additional, potent option. Importantly, Cisco plans to make this enhanced capability available to users under their existing licenses, meaning no additional cost for customers seeking to bolster their AI security posture.
IBM is also set to benefit from Mythos's capabilities. The tech giant will integrate the Anthropic model into its Concert tool, an AI-powered application designed to scan for risks and weaknesses. IBM stated that Mythos will be instrumental in uncovering vulnerabilities that traditional scanners overlook, assessing their real-world risk, and generating potential fixes for developer review.
Anthropic's decision to partner with these industry leaders allows them to distribute Mythos's advanced functionality to a wider audience without granting direct access to the model itself. This controlled rollout is a strategic move, given that a tool as powerful as Mythos could be misused by malicious actors if placed in the wrong hands. Anthropic initially offered limited access through programs like 'Project Glasswing' for vetted organizations, and these new partnerships represent a significant expansion of that strategy.
In parallel with these integrations, Anthropic has also released version 5.2 of Mythos and its companion model, Fable. These updates promise improved performance, with outputs generated up to 30 percent faster than the previous version, and at a reduced cost. Users can opt for a 'fast mode' that further accelerates token generation, albeit at a doubled price point. Anthropic emphasized that extensive safety testing was conducted on these new versions, with a focus on maintaining robust alignment measures and enhancing safeguards against various threats, including distillation attacks, which the company believes are used by some foreign AI labs to clone their models.
The integration of Mythos into Cisco and IBM products underscores the increasing reliance on AI for both offensive and defensive cybersecurity operations. As AI models become more sophisticated, so too does the need for advanced tools to test their security and ensure their responsible deployment. These partnerships signal a maturing market for AI-driven security solutions, where specialized models are being embedded into broader platforms to enhance overall cyber resilience.