AI Research Scientist, FAIR Security, Privacy, and Reliability

Meta; San Francisco (CA); On-site; USD 180,000 - 240,000; Full time

Job summary

Meta's FAIR Security, Privacy, and Reliability team is seeking researchers to break and fix agents and foundational models, driving safer and more reliable AI systems. You will develop benchmarks and contribute to Evaluation Reports, and design post-training mitigations for risks in foundation models.

Collaborate with cross-functional teams to translate research into production systems and publish findings in top venues, advancing measurement of memorization, privacy, and reliability in AI

Qualifications

• PhD preferred or equivalent practical experience.
• Track record of publications in top-tier venues in safety, security, privacy, or robustness of AI models.

Responsibilities

• Conduct fundamental research to discover safety and security failures and robust evaluation methods for AI agents and foundation models.
• Develop benchmarks and contribute to Meta’s evaluation suite and reports.
• Design and implement post-training mitigations for safety, security, privacy, and reliability risks in foundation models.
• Collaborate with cross-functional teams to translate research into production systems.

Skills

Safety research

Adversarial ML

Robustness

Research publications

Education

Bachelor's degree in CS/EE or equivalent

PhD in CS/ML or related field

Job description

About

The FAIR Security, Privacy, and Reliability team breaks and fixes agents and the foundational models that power them. We conduct fundamental research to discover novel attacks, measure privacy, and find non-intuitive breaks in robustness. We frequently contribute those benchmarks to Meta's model Evaluation Reports and have published many of them in top research venues, earning top-1% recognition at ICLR and ICML. The team also lands novel post-training mitigations for these risks in both the open-source line of models and has multiple opportunities for direct research-to-production.

Responsibilities

• Conduct fundamental research to discover novel safety and security failures, robust approaches to measuring memorization risk or reliability failures in AI agents and foundational models,
• Develop and contribute benchmarks to Meta’s foundational model evaluation suite and/or the Evaluation Reports
• Design and implement novel post-training mitigations for safety, security, privacy, and reliability risks in foundation models
• Collaborate with cross-functional teams to translate research into production systems

Minimum Qualifications

• Bachelor's degree in Computer Science, Computer Engineering, relevant technical field, or equivalent practical experience
• PhD in Computer Science, Machine Learning, or related field, or equivalent practical experience
• Research experience in at least one of the following areas: safety, security, privacy, or robustness of AI models; adversarial machine learning; indirect prompt injections or jailbreaks; contextual integrity or memorization; reward hacking or other agent reliability failures; testing or mitigating foundation models for catastrophic risk (CBRNE, cyber, loss of control); or developing mitigations in any of these areas Track record of publications in top-tier research venues