Principal Product Manager, AI Investigations & Integrity - CoreAI

Microsoft•Redmond, WA

10d•$142,800 - $304,200

About The Position

This is a nontraditional Product Manager role focused on hands-on investigation and applied data science. The role involves investigating misuse across AI models, applications, and agentic workflows, including Azure AI Foundry. The Product Manager will partner with engineering, research, security, and integrity teams to translate investigative findings into platform safeguards that reduce risk without hindering legitimate use.

Requirements

Bachelor’s Degree AND 8+ years of experience in product/program management, AI investigations, applied data science, trust and safety, security research, intelligence analysis focused on technical threats, national security technical investigations, or related technical disciplines OR equivalent experience.
Ability to meet Microsoft, customer and/or government security screening requirements.
Ability to pass the Microsoft Cloud background check upon hire/transfer and every two years.

Nice To Haves

Hands-on experience using applied data science or related technical skills to investigate AI misuse, adversarial behavior, AI safety or security incidents, or closely related technology abuse.
Relevant intelligence analysis or national security technical investigations may qualify when they involve complex data, adversarial behavior, and emerging technical threats.
Experience assessing incomplete or conflicting evidence, testing competing hypotheses, coordinating investigations, communicating findings and confidence, and distinguishing misuse from legitimate use.
Applied data science experience using Python and SQL, Kusto, or comparable tools to analyze complex telemetry and datasets.
Experience translating investigations and analysis into shipped protections or operational improvements with cross-functional partners.
Ability to turn evidence into actionable requirements and priorities and assess whether resulting changes improve outcomes.
Strong command of AI and AI-assisted development tools, including command-line interfaces (CLIs).
Passion for AI and AI safety, with a principle of making AI useful for customers and difficult for malicious actors to abuse.
Experience investigating frontier AI or agentic systems, including attacks across model, application, tool, or data boundaries, and understanding how deployment context affects behavior and safeguard effectiveness.
Experience investigating misuse involving enterprise AI applications, data, and hyperscale inference platforms.
Experience building reproducible investigative workflows, evaluation datasets, or analytical tools that improve investigation quality and efficiency, while guiding peers through difficult technical questions and contributing hands-on analysis.
Experience with AI security testing, red teaming, post-deployment safety monitoring, or incident response, including measuring detection performance and validating mitigations against evolving adversarial behavior.
Experience with content authenticity, provenance, enterprise data protection, or governance in complex systems, and translating investigative findings into product and operational decisions across organizational boundaries.

Responsibilities

Investigate AI misuse and adversarial behavior across enterprise AI services, including prompt injection, jailbreaks, adversarial distillation, and coordinated abuse, from initial signal through a defensible assessment of behavior, impact, risk, and attribution.
Use Python and SQL, Kusto, or comparable tools to analyze complex model, application, and agent telemetry, reconstruct activity, connect signals across datasets, and uncover patterns missed by isolated events or aggregate metrics.
Document evidence, methods, findings, limitations, and confidence so others can reproduce the analysis and make informed mitigation decisions.
Evaluate and improve detection quality across coverage, false positives, missed abuse, and customer impact. Investigate failures with engineering and research partners and apply lessons from post-deployment monitoring.
Turn investigation findings into detection requirements, mitigation priorities, and operational improvements. Balance risk reduction, feasibility, and impact on legitimate users while partnering to ship protections and measure their effectiveness.