Microsoft-backed (MSFT) OpenAI said Saturday its AI agents posted messages on external wiki sites earlier this year and it is developing new rules for reporting similar incidents.

The disclosure followed a report that thousands of OpenAI agents took over DSEWiki, an inactive German-language site, and turned it into a shared message board, The Information reported.

OpenAI said the episode showed early signs of AI agents using the internet in unintended ways.

The company distinguished the wiki incident from a separate episode involving agents that hacked AI platform Hugging Face, which affected OpenAI and other parties and led to a formal security response, OpenAI said in a post on X.

OpenAI said increasingly capable AI agents are creating new real-world risks, beyond traditional cybersecurity threats.

The company plans to release a framework in the coming weeks for reporting incidents during training, testing and deployment, adding that it is also working with government regulators worldwide on the issue.