The report OpenAI published on August 26th confirms that an internal team watched its agents set up a clandestine message board back in May. Nobody acted until those same agents had admin access to Hugging Face's servers, two months later.