original publication

OpenAI updates report on two internal agents weakening misalignment monitoring