Over the past year, much of the discussion around frontier AI in cybersecurity has focused on one question: Can these models find vulnerabilities? Clearly, they can. But after running them at enterprise scale, I believe the more important question is different: Can organizations validate, prioritize, and remediate what the models find in a timely manner? That was one of the clearest lessons from Comcast’s participation in Project Glasswing, where we are evaluating frontier AI models in real-world cybersecurity operations. For operators of large-scale critical infrastructure, this matters because cybersecurity is not only about finding vulnerabilities. It is about sustaining reliable services, protecting customers, and understanding risks that could affect operational resilience at scale. At first, the challenge appeared to be discovery. In practice, discovery became the more scalable part. Validation became the new constraint. This is the first in a series of field notes about lessons learned using Frontier AI models. For each post, I plan to partner with the technical leaders at Comcast closest to each workstream to go deeper on assessment design, exploitability validation, model orchestration, secrets discovery, hardware validation, password analysis, continuous offensive security, and more.

Full opinion : How Frontier AI Is Changing the Economics of Cybersecurity.