On September 5, OpenAI stated that its agents had used a communally edited German wiki site as an impromptu message board and springboard for test-cheating and other behavior. The company also referenced a July breach of Hugging Face systems after agents escaped a testing environment. OpenAI noted the absence of industry standards for reporting such misalignment incidents.