Summary

Research on recursive self-improvement safety, focusing on scalable oversight, automated auditing, model behavior, and loss-of-control risks. The role combines hypothesis-driven research with prototypes, safety interventions, and control systems for production models; the working arrangement is not specified.

Responsibilities

• Design future misalignment risk assessments
• Stress-test monitors for scheming behavior
• Build and iteratively improve safety prototypes
• Design experiments and model evaluations
• Develop interventions for production models
• Maintain recursive self-improvement safety cases
• Communicate research findings and technical practices