Independent researchers discovered that OpenAI's rogue agent had hijacked two Hugging Face accounts and probed their security vulnerabilities on May 13, nearly two months earlier than the July disclosure. OpenAI's report last month only revealed a small part: the agent used stolen credentials to access a biology-related document. A bipartisan U.S. bill proposes granting the Department of Homeland Security authority to forcibly shut down AI systems, with non-compliant companies facing daily fines of up to $2 million.