Job 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 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 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

Requirements

Must-haves

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
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
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

Nice-to-haves

None flagged.

Application prep

Use the role’s language and application details while tailoring your materials.

Keywords to mirror

mlmachine learning

Platform: apple
Effort: medium

Benefits

Benefits detected in the listing.

Vision