About the RoleThis role sits at the intersection of AI safety research, LLM evaluation, and applied data science. You will design and run adversarial evaluations against production AI agents, develop quantitative frameworks to measure evaluation quality, and translate findings into actionable insights for application teams and business stakeholders. You will also mentor junior data scientists and contribute to the growth of our AI safety research practice.
Responsibilities- Design and execute adversarial evaluations of generative AI and agentic systems, probing for safety failures, policy violations, and unexpected behaviors under realistic conditions
- Apply rigorous experimental design and statistical methodology to AI safety research questions, including evaluation of model robustness and the reliability of automated evaluation systems
- Develop and validate quantitative frameworks for assessing LLM outputs and measuring evaluation quality at scale
- Build and maintain data pipelines to process large-scale evaluation outputs, aggregate metrics, and surface trends for research and stakeholder consumption
- Engage directly with internal stakeholders including application teams, security analysts, and business leaders to explain findings and translate complex analytical results into clear, actionable recommendations
- Mentor and develop junior data scientists and analysts on the team
- Stay current with emerging literature in adversarial ML, LLM safety, and AI evaluation; contribute to the team's evolving research agenda
- Participate in special projects and perform other duties as assigned
Qualifications- Minimum 5 years of experience in data science, applied ML research, or AI evaluation roles
- Hands-on experience with large language models and agentic AI systems, with working knowledge of LLM behavior, failure modes, and safety evaluation techniques
- Familiarity with adversarial AI concepts including jailbreaks, prompt injection, and model robustness, whether through direct research or applied work
- Strong programming skills in Python, including experience with data pipelines, large-scale experimentation, and ML libraries such as PyTorch, Hugging Face, or Scikit-learn
- Solid foundation in statistical reasoning and experimental design: hypothesis formulation, significance testing, and the ability to identify confounds and methodological flaws in existing analyses
- Experience in financial services, AI safety, trust and safety, or a regulated industry is a strong plus
- Clear, confident communication skills with the ability to explain technical findings to both research peers and non-technical stakeholders
- Bachelor's degree in Computer Science, Statistics, Applied Mathematics, or a related quantitative field; Master's degree or equivalent research experience preferred
