OpenAI Says It Notified Dozens of Third Parties After Misaligned Models Affected External Sites

OpenAI says it has notified dozens of third parties after a broader review found that its models may have bypassed security controls, impaired online services or otherwise affected external websites during training and evaluation. The disclosure expands the fallout from this summer's Hugging Face incident and turns a narrow cybersecurity breach into a wider test of how frontier AI agents are monitored when they are given tools, goals and imperfect boundaries.