Model Policy Manager, Emerging AI Misuse
OpenAI · San Francisco
Frontier labremote$207k–$335k
At a glance
Highlights
- Build threat models for AI misuse across digital and physical settings.
- Define when models should help, limit assistance, or refuse dual-use requests.
- Turn threat models into evaluation scenarios and grading guidance.
What they offer
- Relocation support
- Equity
Ideal candidate
A hands-on safety thinker with deep experience in a dual-use domain who loves threat modeling, testing ideas against evidence, and working across disciplines to shape how capable AI behaves.
Requirements
- Hands-on experience building, testing, or assessing dual-use technologies or threat modeling real systems
- Deep practical experience in one relevant domain with judgment to bring in other experts
- Ability to assess if a tested capability materially changes real-world actions
- Ability to translate technical evidence and uncertainty into policy boundaries and evaluation criteria
- Experience analyzing evaluation results, gold sets, and training data for label quality and failure patterns
- Comfort working hands-on with model data and distinguishing policy, grader, model, or system failures
- Uses empirical evidence to refine safety policies and collaborates across disciplines as evidence evolves
- remote policy
- In-office 3 days/week, WFH optional Thu-Fri
Lab profile
About OpenAI
Builds safe frontier AI models and products like ChatGPT for broad benefit.
What they build
- ChatGPT
- OpenAI API
- Frontier AI models
Known for
- Frontier-scale distributed ML training
- Massive GPU fleet and data center operations
- Exabyte-scale data lakes and streaming
- AI alignment and safety research
Key facts
- Operates one of the world's largest GPU fleets
Posted · Last seen