OpenAI will regularly release its model safety evaluation results.

On May 14th, local time, OpenAI announced the launch of a security evaluation center for OpenAI models. OpenAI stated that as an important measure to actively enhance security transparency, the center's content will be regularly updated. These results only reflect a partial dimension of OpenAI's security work and are intended to provide periodic snapshots. For a comprehensive evaluation of model security and performance, this center's evaluation data should be combined with system card descriptions, prepared framework assessment reports, and specialized research reports released with each model.