Trust and Safety Intel Analyst, Model Behavior and Alignment

Google•Washington, DC

8d•$141,000 - $205,000

About The Position

As a part of the Intel Analysis team, you will identify, analyze, and prioritize threats across a broad surface, deceptive behavior, emerging model capabilities, fraud, global events, violent extremism, and system vulnerabilities. In this role, the analysts combine research, internal signals, and cross-functional expertise to surface emerging trends, forecast risks, and convert outputs into machine-readable data for real-time intelligence integrations. As a part of Trust and Safety operations, policy, and enforcement, you will serve as an intelligence partner across Google. Your work directly shapes how leadership and product teams respond to fast-moving threats, from election crises to novel abuse patterns. In this role, as an Intelligence Analyst, you will shape our understanding of model behavior risks evaluating how AI systems behave in unexpected or misaligned ways, as distinct from intentional misuse by bad actors. In this role, you will own the risk portfolio for model behavior across chatbots, multimodal assistants, agents, and robotics. You will drive collection requirements, synthesize incidents and eval results into clear risk assessments, and distinguish genuine misalignment from misattributed failures or hype. Your insights will directly inform launch readiness, safety cases, and executive decisions, working with safety researchers and policy teams to implement effective mitigations. In this role, you will sit at the intersection of AI safety and threat intelligence as strong candidates bring context in AI safety research, evaluations, and threat modeling, along with sound analytic judgment to evaluate evidence. We welcome applicants from AI safety, policy, intelligence, or Trust and Safety backgrounds.

Requirements

Bachelor's degree or equivalent practical experience.
7 years of experience in one or more of the following: AI safety or alignment research, AI policy or governance, model evaluations or red teaming, trust and safety, intelligence or threat analysis, or risk analysis.
Experience reading and critically assessing technical AI research or evaluation results, and communicating findings to non-specialist audiences.

Nice To Haves

Experience in AI safety, alignment, model evaluations, and AI governance, including familiarity with key labs and evaluators.
AI evaluation experience across modalities, technical fluency to assess risks, and scripting skills (Python/SQL) for analysis.
Deep knowledge of misalignment research (e.g., deception, evaluation awareness, reward hacking) and ability to evaluate evidence.
Strong understanding of AI safety policy, regulatory landscapes, and frontier safety frameworks.
Proven ability to lead complex, multi-quarter projects and influence stakeholders without direct authority.
Background in threat modeling and intelligence generation under uncertainty, including AI-specific adversarial risks like sandbagging.

Responsibilities

Maintain current, company-wide view of material risks across products and modalities, mapping each to owners, mitigations, and remaining gaps, run regular cross-functional reviews of signals from investigations, evaluations, red teaming, user reports, external research, public incidents, and the policy landscape.
Connect incidents and weak signals to broader trends, distinguish genuine misalignment from misattributed failures and hype, and produce quick-turn and strategic assessments, with clear confidence levels and recommendations, for a variety of stakeholders.
Assess how new capabilities and deployment surfaces affect risk, contribute real-world evidence to safety cases, escalate unowned or under-mitigated risks, and track residual risk after launch.
Bring real-world signals into their priorities and turn their findings into taxonomies, threat models, and mitigations with clear owners.
Build AI-assisted workflows that automate research monitoring and signal triage, so analyst time concentrates on judgment-heavy work.