OpenAI Updates Responsible Scaling Policy with Capability Thresholds
Action Required
Failure to adhere to the updated Responsible Scaling Policy could result in increased risk assessments and potentially stricter safeguards for OpenAI's AI models.
AI Impact Summary
This announcement details an updated Responsible Scaling Policy (RSP) from OpenAI, reflecting a year of experience and incorporating learnings from other high-consequence industries. The key change is the introduction of Capability Thresholds – specific AI abilities that trigger stronger safeguards, such as ASL-4 standards for autonomous AI research and ASL-3 standards for CBRN weapons assistance. This demonstrates a proactive approach to managing escalating AI risks and highlights OpenAI’s commitment to proportional protection based on model capabilities, aligning with biosafety levels. The policy also emphasizes continuous monitoring, evaluation, and external input to ensure ongoing effectiveness.
Affected Systems
- Date
- Date not specified
- Change type
- policy
- Severity
- high