Join us as we build safe and reliable foundation models for Apple's products. Our team focuses on the research and methods needed to make large language models safe, fair, and robust at scale, with a current focus on agentic safety, model character and persona, and multimodal safety across audio and vision. We are looking for a researcher with a strong track record in applied ML who cares about safety and responsibility in foundation models. In this role, you will lead research and application of methods that keep Apple's foundation models safe while meeting Apple's standards for privacy and quality. ## Description This role works across many teams. You will work closely with ML researchers, engineers, and product teams to build and ship safety solutions for Apple's foundation models. The most useful research often comes from combining new methods with real product needs, and that is where we focus. As part of this role, you will: \ Define and deliver safety methods for foundation models, with emphasis on agentic safety, model character and persona, and multimodal safety (audio and vision) \ Research the safety and security risks of agentic systems, including tool use, and build the data and evaluations needed to address them \ Help define and safeguard the model's character and persona, so it behaves consistently and safely across contexts \ Build ways to train and evaluate foundation models with safety in mind, including for audio and vision modalities \ Research and improve safety alignment and robustness methods for large language models \ Work with engineers and researchers across teams to improve Apple products \ Partner with product teams to define requirements and drive quality and delivery ## Minimum qualifications A strong ML background with hands-on experience training models, and a record of results shown through publications (e.g., ACL, CHI, CVPR, EMNLP, FAccT, ICML, Interspeech, NeurIPS, UIST) or other clear evidence of impact Strong coding skills, including the ability to work with messy, real-world codebases If not already experienced with LLMs, must be able to quickly pick up the skills needed for safety post-training Strong collaboration and communication skills Willingness to work with sensitive content, including exposure to offensive or controversial material ## Preferred qualifications BS, MS, or PhD in Computer Science, Machine Learning, or a related field, or equivalent experience Direct experience with LLMs and post-training methods, including safety alignment Experience in one or more of: agentic safety (tool use, multi-step agents), model character and persona design, or multimodal safety (audio and vision) Strong organizational skills and experience working across large, diverse teams