IC Technology & Surveillance — 2026-05-08

Pentagon CTO Touts AI Cyber Defense Tools as Anthropic Designated Supply Chain Risk

At the SCSP AI+Expo on May 7, Pentagon CTO Emil Michael and Cyber Policy chief Katherine Sutton said AI models like Mythos could patch vulnerable code in "minutes to seconds" versus the current "weeks to days," Breaking Defense reported. Michael called Mythos "really just an example of the upcoming evolution of cyber-capable models" and said OpenAI, xAI, and Google would have comparable cyber capabilities "in the next year or so." He acknowledged the Pentagon had been "over-reliant on Anthropic products" and pledged it would "never again" be "single-threaded with any one model," Nextgov/FCW reported. Michael cited last week's announcement clearing eight AI firms for classified network deployment as proof the department was moving away from single-vendor reliance.

Analysis
Michael's on-record acknowledgment of over-reliance, corroborated across Breaking Defense, DefenseScoop, and Nextgov/FCW, converts a quiet procurement adjustment into a stated policy commitment with institutional accountability. Eight-firm classified-network clearances already granted create bureaucratic and contractual friction against reconcentration, making it likely the department maintains meaningfully distributed vendor relationships over the next 12 to 18 months. His framing of Mythos as illustrative of capabilities he expects from OpenAI, xAI, and Google within a year signals competitive supplier positioning, not acute gap management. The "never again" language may instead primarily be a negotiating lever against Anthropic pricing. Cyber performance claims about code-patching in minutes remain unverified and carry credibility risk under adversary pressure.

3 sources

Top Pentagon tech officials optimistic Mythos-style AI tools will improve cyber defense - Breaking Defense
Amid concerns sparked by Mythos, the Pentagon's cyber policy chief sees 'huge opportunity' with frontier AI models - DefenseScoop
Pentagon will 'never again' rely on a single AI provider, official says - Nextgov/FCW