Third-party cyber evaluations involving OpenAI models

OpenAI has released third-party cyber evaluations involving its models, detailing how external researchers assessed the models' potential for cyber misuse. The evaluations, conducted by organizations such as the U.S. Department of Defense and the U.K. AI Safety Institute, found that while the models can assist with cyber tasks, they do not significantly lower the barrier for sophisticated attacks. OpenAI stated that the results will inform its safety policies and model deployment decisions.