Security research became agentic

OpenAI’s Aardvark announcement described a security researcher that could help find and fix vulnerabilities at scale. The significant shift was not that a model could generate exploit-shaped text. It was that security work was being organized as a loop of discovery, triage, explanation and coordinated disclosure.

That loop has to preserve the distinction between finding a weakness and safely handling it. An agent needs narrow access, durable evidence and a disclosure process that keeps defenders ahead of attackers. Security automation is useful only when its authority is more constrained than its curiosity.