Third-Party Cyber Evaluations Involving OpenAI Models
OpenAI, Tuesday, August 4th, 2026
OpenAI explains recent third-party cyber evaluation incidents and the safeguards it is adding.
OpenAI addresses recent incidents that occurred during third-party cybersecurity evaluations involving its models.
The post explains what happened in those evaluations and how the situations arose. OpenAI then outlines new safeguards intended to strengthen AI model testing and evaluation procedures.
The disclosure responds to concerns about evaluations producing real-world effects beyond their intended scope.