Researcher, Cyber Safety Training

OpenAI•San Francisco, CA
•$380,000 - $500,000

About The Position

The Cyber Safety Training Researcher will work on large training runs that shape OpenAI’s frontier models, teaching models to support legitimate security work while recognizing when assistance could enable harm. The role studies both over-refusal that can hinder defenders and unsafe assistance, then uses those findings to improve subsequent training runs. The position combines hands-on machine learning and cybersecurity judgment. It may suit someone with cybersecurity experience seeking deeper expertise in post-training, or someone with reinforcement learning experience who wants to build substantial cybersecurity expertise. The researcher works with cybersecurity researchers and the core post-training team, and collaborates across functions to help ensure product safety.

Requirements

  • Hands-on machine learning experience, including having trained or fine-tuned models and being able to describe an experiment whose results informed how a subsequent model was trained.
  • Ability to build and debug code for a training experiment, from preparing data through checking whether observed results reflect a real change in model behavior.
  • Enjoyment of working through ambiguous cases with domain experts and translating findings into concrete decisions about model responses.

Nice To Haves

  • Practical security judgment gained through industry work, a research project, collaboration with a security team, or capture-the-flag competitions.
  • Reinforcement learning experience and motivation to develop cybersecurity judgment alongside security researchers.
  • Cybersecurity experience and interest in developing deeper expertise in post-training.

Responsibilities

  • Work with cybersecurity researchers to identify situations where context changes what assistance a model should provide.
  • Design complex and realistic cyber safety evaluations.
  • Research post-training techniques to robustly improve the cyber safety of frontier models.
  • Work with the core post-training team to incorporate promising methods into large model runs and verify that safety improvements hold at scale.
  • Coordinate and collaborate with cross-functional teams, including Trust & Safety, legal, policy, and other research teams, to ensure products meet high safety standards.
© 2026 Teal Labs, Inc
Privacy PolicyTerms of Service