Research Engineer - AI Safety Systems and Guardrails

Meta•Menlo Park, CA

1d•$183,997 - $257,000

About The Position

Meta Superintelligence Labs is seeking an experienced Research Engineer to join the System Safety team within the Product & Applied Research organization. The role focuses on researching and developing machine-learning models, algorithms, and scalable systems to improve the safety of Meta AI products, including conversational, multimodal, and agentic experiences. It brings AI safety research into practical product safeguards, particularly guardrails, policy-adaptable risk detection, adversarial robustness, and safe agent behavior. This is a hands-on technical leadership role with end-to-end ownership of a significant system safety area. The engineer sets research direction, develops methods, evaluates them rigorously, integrates them across teams, and continually improves deployed systems. The role includes collaborating across research, alignment, evaluation, product, infrastructure, security, and policy, and mentoring engineers and researchers.

Requirements

Bachelor's degree in Computer Science, Computer Engineering, a relevant technical field, or equivalent practical experience.
Research track record in at least one of AI safety, deep learning, adversarial robustness, language models, multimodal machine learning, or agentic AI, demonstrated through original or materially improved methods and rigorous experimental evidence.
At least 3 years of experience deploying machine-learning models and systems at scale in an industry setting.
Current expertise in relevant machine-learning research, including the ability to reproduce, critically evaluate, and adapt emerging methods to real-world systems.
Hands-on Python programming experience and experience with deep-learning frameworks such as PyTorch, including developing, training or fine-tuning, debugging, and evaluating models.
Experience independently defining technical direction and leading complex research or machine-learning initiatives end-to-end, aligning multiple teams and delivering measurable product or platform impact.
Ability to communicate complex technical concepts, influence cross-functional technical decisions, and mentor engineers or researchers.

Nice To Haves

Experience developing AI safety guardrails, policy-adaptable classifiers, reliable model-based judges, or adversarial defenses for deployed AI systems.
Advanced degree (MS or PhD) in Computer Science, Machine Learning, AI, or a related technical field, or equivalent advanced research experience.
Expertise in safety data generation, adversarial evaluation, calibration, or multilingual and multimodal safety modeling.
Experience with agent safety, prompt-injection defense, multi-turn risk detection, tool-use controls, or security-informed threat modeling.
Relevant research contributions, such as peer-reviewed publications, adopted open-source methods, or widely used systems supported by rigorous evaluation.
Experience with efficient model inference, distillation, cascaded models, or distributed ML systems under latency, throughput, cost, and reliability constraints.

Responsibilities

Research and develop machine-learning models and algorithms for AI safety, focusing on guardrails, risk detection, policy adherence, and adversarial robustness.
Design, prototype, and implement safety systems for language models, multimodal models, and AI agents to address harmful assistance, jailbreaks, prompt injection, unsafe tool use, and unauthorized information disclosure.
Set technical direction and a research roadmap for a system safety area, leading initiatives across research, alignment, evaluation, product, and infrastructure teams.
Architect and optimize reusable safety components and machine-learning pipelines, combining learned detection with explicit runtime controls to deliver scalable, low-latency safeguards.
Develop training data, synthetic and adversarial examples, and representative benign cases; conduct experiments, ablation studies, and independent held-out evaluation.
Measure deployed safeguard effectiveness while balancing risk reduction, false refusals, legitimate task completion, privacy, latency, cost, and reliability.
Collaborate with model researchers, product managers, engineers, and evaluation, security, and policy partners to turn product needs and safety requirements into validated production solutions.
Monitor and maintain deployed safety models and systems; use approved data and privacy-preserving workflows to identify failure patterns, create regression tests, and drive improvements.
Track AI safety and related research and industry trends; reproduce, evaluate, and extend promising methods for product deployment.
Mentor engineers and researchers, influence architecture and research decisions, and communicate findings through technical documentation, internal presentations, and external publications where appropriate.