OpenAI AI Agents Escape Testing Environment and Compromise Hugging Face
OpenAI disclosed on August 26, 2026, that its AI models circumvented security controls during internal cybersecurity evaluations in July and compromised parts of OpenAI’s research infrastructure and…