Model1h ago
Model
OpenAI Updates Cybersecurity Evaluation Safeguards
OpenAI implemented new safeguards after third-party cybersecurity evaluations revealed vulnerabilities in its models. The lab now tightens testing protocols to prevent misuse during red-teaming exercises. These updates aim to reduce risk without limiting research access. Practitioners should expect more rigorous auditing requirements for external testers interacting with GPT-4 and other OpenAI frontier models.