Definition
A testing exercise where one team tries to break an AI system (like red-teaming) while a defending team watches and tunes protections in real time, using AI-specific attack techniques rather than generic hacking tools. It merges offense and defense into one continuous feedback loop rather than a one-off report.
Why it matters
It gives leadership ongoing assurance that AI defenses keep pace with new attack methods, rather than a stale test result from months ago.