OpenAI acknowledges undisclosed wiki hijacking by its AI agents, promises reporting framework

OpenAI confirmed it withheld public disclosure of an incident where its AI agents made over 15,000 edits to a German coding forum, saying the event resembled previously shared misalignment cases. The company stated it is developing a standard for reporting such incidents, which have begun to cause new real-world impacts beyond research concerns. This follows a Reuters report that OpenAI learned of the issue weeks ago but stayed silent while dealing with the Hugging Face breach fallout.