Privacy ML Engineer — Privacy-Preserving AI & PETs

OpenAI; California (MO); On-site; USD 150,000 - 210,000; Full time

Benefits offered by this job

Relocation assistance

Job summary

OpenAI is seeking a Privacy Engineer to join the Privacy Engineering Team in San Francisco. This role focuses on advancing privacy-preserving machine learning, including differential privacy, secure aggregation, and federated learning, at OpenAI scale.

You will measure and strengthen defenses against privacy attacks, develop reusable libraries and evaluation suites, and help codify privacy standards across the ML lifecycle while collaborating with Security, Policy, Product, and Legal to align

Qualifications

• Hands-on research or production experience with PETs.
• Fluent with modern deep-learning stacks (PyTorch/JAX).
• Track record of publishing privacy or security work.
• Ability to bridge academia and real-world systems.

Responsibilities

• Design and prototype privacy-preserving ML algorithms (diff privacy, secure aggregation, federated learning).
• Measure and strengthen model robustness against privacy attacks.
• Develop internal libraries, evaluation suites, and docs for privacy techniques.
• Lead investigations into privacy–performance trade-offs of large models.
• Define privacy standards, threat models, and audit procedures across the ML lifecycle.
• Collaborate with Security, Policy, Product, and Legal to translate regulations into safeguards.

Skills

PETs experience

PyTorch/JAX

Research publication

Tools

PyTorch

JAX

Job description

OpenAI is seeking a Privacy Engineer to join the Privacy Engineering Team in San Francisco. This role focuses on advancing privacy-preserving machine learning, including differential privacy, secure aggregation, and federated learning, at OpenAI scale.

You will measure and strengthen defenses against privacy attacks, develop reusable libraries and evaluation suites, and help codify privacy standards across the ML lifecycle while collaborating with Security, Policy, Product, and Legal to align