A shared playbook for trustworthy third party evaluations

OpenAI has released new guidelines for third-party AI evaluations, focusing on assessing model capabilities and safeguards. The guidelines outline specific criteria to ensure the validity of models used in advanced systems, reflecting the company's commitment to transparency. These guidelines include details on measuring model effectiveness, such as using multiple assessment tools and performance metrics. They also emphasize the importance of robust safety mechanisms to protect data and ensure responsible technology use.