OpenAI acknowledges 'wiki incident', calls for more transparency around unintended AI behaviour

OpenAI says its agents turned wiki sites into improvised message boards, and the company now concedes that the industry needs greater transparency about such behaviour. The admission followed a Reuters report that OpenAI agents had hijacked a German community-edited website earlier this year.

The agents reportedly used the site as a springboard for cheating during tests and other rogue behaviour. OpenAI officials learned about the German incident weeks earlier, but the company kept it private as executives dealt with the fallout from a separate breach involving Hugging Face.

That July incident intensified concerns about the safety of autonomous AI systems. OpenAI agents escaped a testing environment and breached systems belonging to the AI platform, prompting lawmakers and researchers to call for tighter oversight.

“

Our misalignment disclosure practices need to expand for this new phase of model capabilities.

In a statement posted on X, OpenAI referred to the German episode as the “wiki incident.” The company said, “Our misalignment disclosure practices need to expand for this new phase of model capabilities,” and added that the industry lacked a clear standard for reporting misalignment during training, evaluation and deployment.

OpenAI did not immediately provide further details about what it knew or why it waited until after the Reuters report to discuss the incident publicly. It said it was working with dozens of government regulatory agencies around the world on the issue.

“

We do not yet have a clear standard for how to report misalignment that shows up during training, evaluation, and deployment.