TechFlow news: OpenAI has released a beta version of a new set of rules called the "Preparedness Framework," designed to ensure the safety of frontier artificial intelligence models.
Under this framework, OpenAI will assess and continuously update a "scorecard" for all frontier AI models.
In addition, OpenAI will set risk thresholds that trigger baseline safety measures, defining four levels of safety risk. Only models with post-mitigation scores at "medium" or below may be deployed, while models must achieve post-mitigation scores at "high" or below to proceed with further development. For models classified as high-risk or critical-risk, OpenAI will implement additional safety measures.
Notably, OpenAI stated it is establishing a cross-functional safety advisory group to review all reports, which will be submitted simultaneously to both leadership and the board of directors. While leadership retains decision-making authority, the board holds the right to override decisions.
