OpenAI
Researcher, Multimodal Safety
$380K–$500K
San Francisco · Full-time · Posted Aug 3, 2026
About the job
Research and develop training and evaluation methods that improve the safety of multimodal models across text, vision, and audio.
What you'll do
- Advance multimodal safety research connecting perception and semantic understanding to safe behavior.
- Build VLM training, post-training, safety evaluation, and intervention methods.
- Partner with research and product teams to apply findings to ambient, embedded, and personalized multimodal experiences.
What they ask for
- Track record building or advancing multimodal models, with depth in a relevant modality or multimodal reasoning and fluency in perception and language.
- End-to-end understanding of multimodal systems, including encoders, projection layers, modality fusion, cross-modal reasoning, scaling, and inference tradeoffs.
- Experience improving frontier model behavior through post-training approaches such as SFT, RL, data curation, synthetic data, evaluation, and error analysis.
- Strong research and engineering judgment for formulating hypotheses, designing experiments and evaluations, diagnosing failures, and developing improvements.
Pay
$380K – $500K USD base salary
Tools
VLMs, SFT, RL
AI focus
Multimodal safety research and safer model behavior across text, vision, and audio, including VLM post-training and evaluations.
Model-written summary · the original posting is authoritative