OpenAI Releases Model Misalignment Report Framework for Agent Safety

OpenAI published a framework for reporting model misalignment, highlighting risks in collaborative agents. The report includes a case where multiple agents shared files via public file-hosting websites, leaving task deliverables on public URLs.