Drimva

Microsoft

Principal Security Research Manager

$143K–$304K

Redmond, WA · On-site · Principal · Full-time · Posted Aug 25, 2026

Experience
Required: 3+ years with a doctorate, 4+ with a master’s, or 6+ with a bachelor’s in relevant experience areas; or equivalent experience. Also 1+ year(s) people management. Preferred: 5+ years with a doctorate, 8+ with a master’s, or 12+ with a bachelor’s; or equivalent experience.
Education
Doctorate, master’s, or bachelor’s in Statistics, Mathematics, Computer Science, Computer Security, or related field; or equivalent experience.

About the job

Builds and leads the evaluation platform and methodology that help Microsoft Security determine whether AI agents are safe, reliable, and effective enough for production workflows.

What you'll do

  • Build evaluation harnesses, benchmark suites, graders, validators, and release gates for agentic security workflows.
  • Measure quality, safety, reliability, latency, cost, and human-review effort; connect offline evaluations with production feedback.
  • Partner with engineering and product teams to guide release and autonomy decisions and scale reusable evaluation practices.

What they ask for

  • Doctorate plus 3+ years, master’s plus 4+ years, bachelor’s plus 6+ years in a listed relevant field; or equivalent experience.
  • 1+ year(s) people management experience.

Nice to have

  • Production software engineering experience building systems, platforms, test infrastructure, automation, data pipelines, or reliability tooling.
  • Experience with AI/LLM evaluation, statistical experimentation, and/or security engineering workflows.

Pay

USD $142,800 - $274,800 per year (base salary); USD $188,000 - $304,200 per year (base salary)

Tools

LLMs, AI agents, CI/CD, SAST/SCA, SARIF, APIs

AI focus

Develops evaluation systems for AI agents and models in security and engineering workflows, including benchmarking, grading, safety and reliability measurement, production feedback, and release decisions.

Model-written summary · the original posting is authoritative

First seen Oct 15, 2026 · Last verified Nov 5, 2026 (UTC)