Safety Post-training (AI)

Location: Redmond, Washington, United States

Salary: USD $119,800–$234,700 per year across the U.S.; USD $160,200–$261,000 per year in the San Francisco Bay Area and New York City metropolitan area.

Company

Microsoft AI develops frontier AI systems and products, including responsible AI, model development, product engineering, and safety infrastructure.

What you will do

• Design and build pipelines for safety data collection, generation, evaluation, experimentation, and model feedback.
• Develop evaluation suites, datasets, metrics, and graders for text, coding, and agentic models.
• Define schemas and APIs for policy taxonomies, annotations, grader outputs, datasets, tool trajectories, and evaluation results.
• Connect production signals, incidents, red-team findings, and partner feedback to training data, evaluations, interventions, and deployment decisions.
• Collaborate with research, engineering, product, policy, security, and partner teams to improve model and system safety.

Requirements

• Bachelor’s degree in computer science, machine learning, statistics, mathematics, engineering, or a related technical field, or equivalent practical experience.
• Strong programming skills in Python.
• Experience training, fine-tuning, evaluating, or researching large language models.
• Experience designing and analyzing rigorous model evaluations, experiments, benchmarks, or measurement systems.
• Experience with large datasets, reliable data or workflow pipelines, data quality, annotation design, and automated graders.
• Ability to translate ambiguous safety or policy questions into measurable model behaviors and communicate results across research, engineering, product, and policy teams.

Nice to have

• Experience with safety post-training methods, reinforcement learning infrastructure, rollout systems, agentic training environments, verifiers, or distributed model training.
• Experience evaluating or mitigating jailbreaks, prompt injection, unsafe tool use, reward hacking, deception, harmful cyber behavior, or over-refusal.
• Cybersecurity expertise in application or cloud security, vulnerability research, penetration testing, secure coding, incident response, identity, or security engineering.
• Experience with statistical methods, human-evaluation design, automated-grader calibration, benchmark contamination analysis, subgroup analysis, distribution shift, or evaluator drift.
• Experience building data-generation, annotation, evaluation, experiment-tracking, or model-lifecycle platforms using workflow orchestration, distributed compute, cloud infrastructure, lineage, governance, or production telemetry.

Culture & Benefits

• Work spans text, multimodal, and agentic AI systems.
• Collaboration with research teams, production inference, security, responsible AI, product organizations, external partners, and customers.
• Benefits and additional compensation may be available depending on the role and work location.
• The position remains open for at least five days, with applications accepted until the role is filled.