The Black Box Play: How Cisco and IBM are Monetizing Anthropic's Mythos

For enterprise software leaders, the primary hurdle in deploying cutting-edge AI isn't just the capability—it's the risk. When a model is powerful enough to find critical software flaws, it is powerful enough to be weaponized. This tension is the core of the new integrations announced by Cisco and IBM involving Anthropic's Mythos model, as The Register first reported.

According to reporting from The Register, Cisco and IBM are integrating Mythos into their respective security suites to enhance bug-hunting and risk assessment. Cisco is incorporating the model into its AI Defense suite, which focuses on identifying weaknesses in AI models, agents, and applications. Jeff Schultz, Cisco's senior vice president for portfolio strategy and marketing, stated that AI Defense has the capacity to "look at models and break them," and that the tool is now "super-charged by Mythos" to identify more danger.

Similarly, IBM is integrating Mythos into its Concert tool. IBM intends to use the model to identify weaknesses that traditional scanners overlook, determine the actual risk associated with those flaws, and generate fixes for developers to approve.

From an operational lens, the real story here is the delivery mechanism. The Register notes that the Cisco deal allows end-users to access the functionality of Mythos without actually using the model itself. This "black box" approach allows Cisco and IBM to offer high-end AI capabilities while avoiding the intellectual property risks associated with a full model deployment.

This controlled access is a strategic necessity. As The Register reports, Anthropic previously managed access to Mythos through "Project Glasswing," a program for vetted organizations, because the model could potentially help attackers exploit software flaws if it fell into the wrong hands. By acting as the intermediary, Cisco and IBM can monetize the model's output without risking the leakage of the underlying IP.

This concern over IP is further highlighted by Anthropic's recent updates. Along with the integrations, Anthropic introduced the 5.2 versions of Fable and Mythos. While these versions offer faster output and specific pricing—$8 per million input tokens and $40 per million output tokens (or double that for "fast mode")—Anthropic explicitly mentioned implementing its "strongest-ever protections against distillation attacks." According to The Register, this is a significant admission, as Anthropic believes Chinese AI labs use distillation to clone its models to avoid expensive training costs.

By leveraging these integrations, Cisco is making the capability available under existing user licenses at no additional cost. Meanwhile, Natalie Mead, Anthropic's head of applied AI for the Asia-Pacific, noted that partnering with Cisco helps the company satisfy a "long queue of buyers" who want to utilize Mythos in a controlled environment.