AIML - Machine Learning Engineer, Red Teaming - Responsible AI and Safety

Written by the T&S Network from Apple's posting.

Red-teaming for Apple's Responsible AI function, probing Apple Intelligence features before and after they ship to find real, reproducible safety vulnerabilities. The team works on attack discovery across features and models, including agentic scenarios, and on tooling to scale that testing.

• AI misuse and jailbreaks

What you'd do

• Own a cluster of Apple Intelligence features and build expertise in their architecture and failure modes
• Probe features pre-ship and post-ship for reproducible vulnerabilities
• Invent novel attack strategies and track newly published jailbreak and attack research
• Check findings for reality, rank them by severity and reach, and push fixes through with stakeholders
• Build automated red-teaming tools that generate attacks, run them in batches, and feed results back into testing
• Identify policy and evaluation coverage gaps and feed them back to eval and mitigation teams
• Work with internal and external crowd and adjacent security teams to route findings

What it asks for

• Hands-on work with generative models for evaluation or product development
• Strong production-quality Python engineering skills
• Deep experience with foundation model programming, such as DSPy prompt optimization
• Experience working with noisy crowd-based labels and human evaluations
• Experience in AI red-teaming (preferred)
• Experience in the Responsible AI space (preferred)

Worth knowing

• Works with highly sensitive content, including offensive and controversial material

Terms in this role

Red teaming.

Testing a system, often an AI model, by deliberately trying to make it fail or misbehave, so problems are found and fixed first.

Jailbreak.

A prompt written to get an AI model to ignore its safety rules.

Our best summary of the posting, and it can be wrong. The employer's posting is what counts: read it in full on Applenorth_east before you apply. If we got something wrong, report this listing.