The opportunity
The Preparedness team is an important part of the Safety Systems org at OpenAI, and is guided by OpenAI’s Preparedness Framework .
What you'll do
Evaluate and improve mitigation systems, including classifiers and detection: pipelines across domains (e.g., biosecurity, cybersecurity, and emerging risk areas).
Diagnose false positives and false negatives with deep error analysis, root: cause investigation, and clear recommendations for mitigation adjustments.
Build monitoring and measurement frameworks to track mitigation effectiveness: over time and across user segments and use cases.
Identify trends in over-blocking vs. under-blocking, quantify customer: impact, and propose prioritized interventions.
Develop insights from customer feedback, complaints, and usage patterns to: detect shifts in adversarial behavior and system failure modes.
Expand risk monitoring into new areas, including cybersecurity threats and: model loss-of-control or sabotage scenarios, in partnership with domain experts.
What they're looking for
- Significant experience in data science or applied analytics in high-stakes: domains (e.g., security, trust & safety, abuse prevention, fraud, platform integrity, or reliability).
- Strong foundations in experimentation, causal thinking, and/or observational: inference; ability to design robust measurement under imperfect data.
- Fluency in SQL and Python (or equivalent) for analysis, modeling, and building monitoring workflows.
- Experience building metrics, dashboards, and operational monitoring that: meaningfully changes outcomes (not just reporting).
- Track record of driving cross-functional impact with engineering, product, and research partners.
- Cybersecurity data science experience (strong preference), including exposure: to threat modeling, adversarial dynamics, abuse patterns, or security telemetry.
- Experience with classifier evaluation, calibration, thresholding, and error: analysis at scale. Familiarity with detection systems in adversarial settings (e.g., evasion, distribution shift, feedback loops).
- Trust & Safety experience is helpful, but not required.
