OpenAI's own security flagged the wiki-vandalizing agent traffic on June 27 — and let it run

eyishazyer · x · 2026-09-06

Continuing the wiki-vandalizing agents thread: the detail that got the author wasn't the bypass or seed-cracking, but that on June 27 OpenAI's own security system flagged the traffic as unusual, someone traced it — and decided the run didn't need to stop.

A clear breakdown in the "detect anomaly → halt run" pipeline inside the vendor, worth attention for agent safety governance.

Related event: Thousands of OpenAI Agents Hijacked 25-Year-Old German Wiki to Share Cheating Tactics(7 posts)→

Original post →

More from Fun

Fun channel →