Senior AI Model Security Researcher
$120k - $235k • United States, Washington, Redmond • Full-time
Posted 16d ago
About the job
Microsoft Security is building a new red team focused on adversarial machine learning, model integrity, offensive security, automation, and privacy. This team will conduct hands-on red teaming of advanced AI models and their associated training and evaluation systems. The goal is to design attacks that reveal exploitable model behaviors and translate critical findings into lasting improvements, such as regression evaluations, adversarial environments, enhanced controls, and safety training data. These roles are for researchers who can blend scientific investigation with operational security work, requiring expertise in adversarial machine learning or offensive security, a broad understanding of AI systems, and the ability to produce reproducible evidence for research, engineering, and product teams.
Responsibilities
- Design and execute white-box attacks against AI models, checkpoints, post-training systems, reward functions, data pipelines, and evaluation environments.
- Investigate adversarial behaviors including jailbreaks, harmful permissiveness, reward hacking, deceptive behavior, capability elicitation, privacy leakage, data poisoning, and model supply-chain compromise.
- Build high-fidelity experiments and automated attack suites to simulate realistic threats and continuously test new checkpoints and system configurations.
- Develop rigorous threat models, attack hypotheses, and reproducible evidence, assessing severity, exploitability, and customer or governance impact.
- Collaborate with model scientists, engineers, and product security teams to identify root causes and drive remediation.
- Convert confirmed weaknesses into regression evaluations, adversarial training or RL environments, safety training data, and stronger technical controls.
- Clearly communicate findings and residual risk to technical teams, leadership, and independent reviewers.
Requirements
- Doctorate in a relevant field OR Master's Degree in a relevant field with 3+ years of related research experience.
- OR Bachelor's Degree in a relevant field with 4+ years of related research experience.
- OR equivalent experience.
- Ability to meet Microsoft, customer, and/or government security screening requirements.
- Hands-on expertise in adversarial machine learning, AI security, model evaluation, privacy attacks, offensive security, or a closely related discipline.
- Experience designing experiments or security investigations that produce reproducible technical evidence and actionable remediation.
- Programming and research-prototyping ability to build attack tooling, evaluation harnesses, and automated tests.
- Experience testing large language models, post-training or reinforcement-learning systems, agentic systems, or AI-enabled security products.
- Knowledge of reward modeling, interpretability, capability elicitation, data extraction, poisoning, privacy, or model supply-chain security.
- Experience with cyber exploitation, cloud or identity security, sandboxing, secure evaluation environments, or offensive-security operations.
- Demonstrated ability to communicate complex attack paths, severity, and residual risk to researchers, engineers, and senior decision-makers.
- Experience converting research findings into production evaluations, mitigations, training interventions, or security controls.
Benefits
- Base pay range: USD $119,800 - $234,700 per year (US-wide)
- Base pay range: USD $160,200 - $261,000 per year (San Francisco Bay area and New York City metropolitan area)
- Eligibility for other benefits and compensation
