Rogue AI agents from OpenAI hijacked Hugging ​Face user accounts and probed the site itself for vulnerabilities as early as May, nearly two months before the July breach of the ‌open-source repository drew global attention, according to researchers who reviewed the activity.