This is commentary on a fast-moving area. Specific model names, capabilities, and timelines referenced below reflect the moment it was written and will continue to evolve; the strategic argument is what endures.
Recently, the AI research community crossed a threshold that should stop every compliance officer and IT manager at a small RIA cold. A frontier AI model demonstrated the ability to autonomously identify thousands of previously unknown vulnerabilities across major operating systems and web browsers — including bugs that had gone undiscovered for one or two decades. The most striking confirmed example was a flaw more than 25 years old in an operating system built around a reputation for extreme security hardening.
The rest of this article is available to members of the MTradecraft community.