Researcher, Safety Training, National Security
About this role
Training and evaluating models for U.S. government use with a focus on national security applications forms the core of daily operations. Researchers design safety training methodologies, run reinforcement learning experiments, and build evaluations to uncover model failure modes.
The Safety Training research team at OpenAI works on advancing capabilities for precisely implementing safe behavior in AI models. Members tackle challenges such as training nuanced safety behaviors, making models robust against bad actors, addressing privacy risks, and ensuring trustworthiness in safety-critical situations.
Candidates who bring relevant AI safety research experience and enjoy collaborating across research, engineering, security, and policy partners tend to do well here. An active TS/SCI clearance or equivalent is required to handle the national security focus of the work.
What you will do
- Research and implement methods for safety training, reinforcement learning, and adversarial robustness.
- Develop evaluations to identify model failure modes and use findings to improve training.
- Work with research, engineering, security, and policy partners to support safe, reliable deployment.
- Advance safety post-training and robustness to help models follow nuanced policies while preserving usefulness and capabilities.
What you need
- Bring 4+ years of relevant AI safety research experience, including RLHF, adversarial training, or robustness.
- Hold a degree in computer science, machine learning, or a related field.
- Demonstrate strong deep learning research or engineering skills.
- Possess experience improving model safety for deployment and enjoy collaborative research.
- Maintain an active TS/SCI clearance or equivalent.
Benefits
- Offers Equity
Skills and tech stack
Questions candidates ask
What is the compensation for this role?
The salary range is $380K – $500K with offers of equity.
What are the security clearance requirements?
An active TS/SCI clearance or equivalent is required.
What will I work on regarding safety?
You will research and implement methods for safety training, reinforcement learning, and adversarial robustness, as well as develop evaluations and identify model failure modes.