Definition
Using AI agents to automatically probe an AI system for weaknesses and exploitable behaviour, rather than relying only on human testers. Frontier labs and vendors are now disclosing that automated red-teaming agents outperform humans and can drive actual exploit execution.
Why it matters
As attackers themselves automate with AI, defenders need AI that can find and test weaknesses at AI speed — the fact that automated red-teaming now beats human testers is both an emerging capability and a warning about how fast the offensive side is moving.