The opportunity
OpenAI's Safety Research organization works to ensure that our most capable AI models can be deployed safely and responsibly. As frontier models become increasingly capable and autonomous, developing reliable ways to monitor their behavior, identify emerging risks, and evaluate…
What you'll do
Lead the research and technical direction for evaluating and improving the: misalignment monitoring of frontier AI systems.
Design and execute rigorous evaluations that measure monitor effectiveness,: identify failure modes, and assess the reliability of monitoring signals.
Develop methods to stress-test and red-team monitors, including identifying: behaviors that evade detection or expose weaknesses in existing approaches.
Investigate false positives, missed detections, and other monitoring failures: to improve the accuracy and usefulness of safety evaluations.
Build scalable evaluation frameworks and experimental tools that enable: ongoing measurement of monitoring performance as model capabilities evolve.
Translate research findings into actionable improvements, working with: partner teams to strengthen monitoring systems and inform broader safety decisions.
What they're looking for
- Establish research priorities and guide technical execution, balancing: foundational research questions with practical deployment needs.
- Have a strong background in machine learning research, particularly in model: evaluations, AI safety, alignment, robustness, or related areas.
- Have demonstrated the ability to lead technically challenging research: projects, from identifying important questions through designing experiments and delivering results.
- Bring strong experimental judgment and understand how to design evaluations: that are rigorous, reproducible, and difficult to game.
