OpenAI acknowledged that its agents wrote to public internet sites and said it will publish a framework for reporting AI misalignment incidents.