Anthropic has announced a strategic expansion of its Mythos AI model, integrating the specialized bug-finding technology into enterprise security platforms managed by Cisco and IBM. This move marks a shift in how the company distributes its high-capability models, moving away from exclusive, vetted programs toward broader commercial availability within established security suites. Under these new arrangements, end-users will gain access to the analytical capabilities of Mythos through third-party interfaces without ever interacting with or hosting the underlying model directly.
Cisco is incorporating Mythos into its AI Defense suite, a platform designed to identify vulnerabilities and inappropriate behaviors within AI models, agents, and their associated applications. By integrating Mythos, Cisco aims to enhance the red-teaming capabilities of its existing tools, allowing customers to stress-test their AI deployments more effectively. Cisco has confirmed that this functionality will be included in existing user licenses at no additional cost, positioning the integration as a performance upgrade to its current automated security testing features.
IBM is similarly adopting the technology, integrating Mythos into its Concert platform. This tool is primarily used to scan enterprise applications for security risks and structural weaknesses. IBM stated that the inclusion of Mythos is intended to identify complex vulnerabilities that traditional scanners often fail to detect. Furthermore, the integration is designed to assist developers by prioritizing which identified risks are most critical and providing automated suggestions for remediation, which developers can then review and approve.
The transition toward these partnerships follows the initial phase of Mythos’s release, known as Project Glasswing. That program restricted access to a small group of vetted organizations to prevent the model’s powerful exploit-generation capabilities from falling into the hands of malicious actors. Anthropic has maintained a cautious approach to distribution, acknowledging that the same features that allow Mythos to identify software flaws could theoretically be repurposed to facilitate cyberattacks if the model were improperly secured or leaked.
Alongside these enterprise integrations, Anthropic has released version 5.2 of both the Mythos and Fable models. The company reports that these updates offer a 30 percent increase in output speed compared to the 5.1 versions, alongside a revised pricing structure. Users now have the option to utilize a fast mode, which accelerates token generation by up to 2.5 times while doubling the cost per million tokens.
Anthropic claims that version 5.2 underwent rigorous pre-release safety evaluations, specifically targeting high-stakes risks such as sandbox escapes. The company asserts that the models remain well-aligned with safety standards and perform as well as, or better than, previous iterations. These evaluations were conducted in-house to ensure that the increased performance did not come at the expense of security or reliability.
A significant focus of the 5.2 release is the implementation of enhanced defenses against distillation attacks. Anthropic has previously expressed concern that foreign AI laboratories may attempt to clone its proprietary models by distilling their outputs. By strengthening these protections, the company aims to prevent unauthorized replication of its intellectual property while continuing to scale the availability of its tools to corporate partners.
While Anthropic has confirmed these partnerships with Cisco and IBM, it has not yet disclosed whether additional third-party integrations are planned. The company continues to manage access to the Mythos model through controlled channels, balancing the demand for high-performance security automation with the inherent risks associated with deploying advanced bug-finding AI.
