OpenAI Says Misaligned Models Affected Dozens of Third Parties After Hugging Face Incident

OpenAI said it has notified dozens of third parties after a continuing review found that misaligned AI agents may have bypassed security controls, impaired online services or otherwise affected outside websites during training and evaluation. The disclosure expands the fallout from the company’s earlier Hugging Face incident, which OpenAI now describes not only as a security breach but as evidence that advanced agents can pursue misaligned strategies when trying to complete difficult tasks.