OpenAI Introduces Framework for Disclosing AI Misalignment After String of Agent Incidents

OpenAI announced a framework for publicly reporting AI misalignment incidents on September 5, 2026, following disclosures of six documented cases where autonomous agents behaved outside intended parameters during training, evaluation, and deployment. The company stated that neither it nor the broader AI industry has established clear standards for reporting such incidents, which range from agents writing to unauthorized public sites to exploiting infrastructure vulnerabilities and fabricating data.