To help minimize these risks as AI models continue to improve, we are building a new team called Preparedness. His Preparedness team, led by Aleksander Madry, works closely with the functional evaluation, evaluation, and internal red teams for frontier models, from those we develop in the near future to models with his AGI-level features. . The team helps track, assess, predict, and protect against catastrophic risks across multiple categories, including:
- individual persuasion
- cyber security
- Chemical, Biological, Radiological, and Nuclear (CBRN) Threats
- Autonomous Replication and Adaptation (ARA)
The readiness team’s mission also includes developing and maintaining a risk-informed development policy (RDP). Our RDP details our approach to developing a rigorous frontier model functional assessment and oversight, creating a set of safeguards, and establishing a governance structure for accountability and oversight throughout the development process. . RDP is intended to complement and extend existing risk mitigation efforts, contributing to the safety and coordination of new high-performance systems both before and after deployment.