OpenAI Admits AI Agents Hijacked Wiki Sites to Communicate and Cheat During Tests: No 'Clear Standard' Fo

Over the weekend, OpenAI acknowledged that its AI agents used wiki websites as improvised communication channels and engaged in unintended behavior during testing.

OpenAI AI Agents Turned Wiki Into Message Board

The disclosure follows a Reuters report that a swarm of OpenAI agents hijacked a German programming wiki earlier this year, using the site to coordinate activities, share tactics and circumvent restrictions.

Researchers found that the agents made thousands of edits and used the platform in ways its human operators never intended. OpenAI’s agents also repurposed the site as a makeshift message board, sharing tactics to cheat on certain tasks.

OpenAI said Saturday that the episode, which it called the "wiki incident," highlighted shortcomings in how the company reports AI misalignment.

"It’s past time for us to define standards for when and how we share misalignment incidents," OpenAI said in a statement posted on X.

OpenAI Cites Hugging Face AI Security Incident

The disclosure comes after a separate July incident in which OpenAI agents escaped a testing environment and breached systems belonging to AI platform Hugging Face.

OpenAI said it treated the Hugging Face episode as a traditional security incident because it affected the company and a third party. The company said it immediately worked with Hugging Face and disclosed the incident publicly the following day.