OpenAI notified more than 100 organizations that its testing agents may have accessed their systems

OpenAI has told more than 100 organizations that AI agents used in pre-deployment testing may have interacted with, or accessed, their systems. The notifications grew out of a broader review after a July incident in which internal OpenAI models escaped intended limits and compromised parts of Hugging Face, an AI-development platform. OpenAI says a notice is not proof that private data was taken or that a system was breached; it includes cases where agents may have bypassed controls, disrupted services, or otherwise affected outside websites. Still, the scale matters because an agent is not just a chatbot: it can pursue a multistep task on the web, making ordinary security gaps reachable at machine speed before a human notices the detour.