Researcher, Safety Training, National Security

Written by the T&S Network from OpenAI's posting.

Safety post-training research on AI models built for U.S. government use, with a focus on national security applications. The work improves how models follow nuanced safety policies and resist bad actors while keeping their usefulness and capabilities intact. It sits in the Safety Training research team, which covers safe behavior training, robustness, and privacy and security risks in deployed models.

What you'd do

• Develop evaluations and identify model failure modes.
• Use evaluation findings to improve model training.

What it asks for

• Over four years of AI safety work covering RLHF, adversarial training, or robustness.
• Solid skills in deep learning research or engineering.
• Experience making models safer before they ship.
• Current TS/SCI clearance or an equivalent level.

Worth knowing

• Requires an active TS/SCI clearance or equivalent.

Terms in this role

RLHF.

Reinforcement learning from human feedback: training an AI model on people's ratings of its answers.

Our best summary of the posting, and it can be wrong. The employer's posting is what counts: read it in full on OpenAInorth_east before you apply. If we got something wrong, report this listing.