OpenAI announced Aardvark, an agent it says can scan code like a security researcher: find issues, check whether they are real, and propose a fix. It is in private beta after internal and partner use.
The company reports high recall on test repositories with known bugs. That is a lab number. What matters in production is false alarms, missed cases, and who reviews the output before a patch ships.
This is the same pattern as other “agentic” tools: useful for volume and a first pass. Judgment — what to fix, what to ignore — stays with a named team.
OriginalOpenAIopenai.comRead the full story