SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
OpenAI's Chat and Multimodal Safety team is seeking a Researcher to advance safety for frontier multimodal AI systems. The team is responsible for ensuring that OpenAI's increasingly multimodal models and products (text, speech, visuals) behave safely across diverse user interactions and emerging harms.
In this role, you will define and advance multimodal safety research across text, vision, and audio modalities, connecting perception and semantic understanding to safe model behavior. You'll build training and evaluation methods for vision-language models, including post-training approaches, safety evaluations, and interventions that help models respond appropriately in varied contexts. You'll collaborate closely with product, model training, and Personal AGI teams to translate research into safer ambient, embedded, and personalized multimodal experiences.
Ideal candidates combine deep technical expertise with strong safety judgment and bridge perception and language. You should have a track record building or advancing multimodal models—with depth in vision-language models, video understanding, image generation, audio, or multimodal reasoning. You understand how multimodal systems work end-to-end, from encoders and projection layers to modality fusion, cross-modal reasoning, scaling, and inference tradeoffs. You've improved frontier model behavior through post-training using SFT, RL, data curation, synthetic data, evaluation, and rigorous error analysis. You bring strong research and engineering judgment to open-ended safety problems: forming testable hypotheses, designing decisive experiments, diagnosing model failures, and translating findings into robust improvements.
This is a hybrid role based in San Francisco, requiring 3 days in the office per week. OpenAI offers relocation assistance to new employees.
