Founding AI Security Researcher

Role overview Investigate security risks in AI agents and turn research findings into reproducible evaluations, practical prototypes, and product guidance. This part-time founding role combines rigorous study of model behavior and policy verification with clear evidence that can inform engineering and research publications.

Responsibilities - Design and assess probabilistic verifiers that can make deterministic policy decisions more cautious. - Build reproducible attack scenarios and evaluations covering prompt injection, tool poisoning, identity spoofing, model misuse, adversarial manipulation, and failures involving multiple agents. - Maintain versioned datasets, experiment records, ablation studies, scorecards, and regression thresholds. - Measure quality alongside latency, reliability, scalability, and cost rather than relying on benchmark scores alone. - Prototype techniques for tracing instruction provenance and attesting that actions were taken independently. - Document experiments, limitations, and reproducibility details for research publications.

Requirements - Research experience with model behavior, agent security, policy verification, or security evaluation. - Ability to formulate a written hypothesis and threat model, then test them with reproducible code and versioned inputs. - Experience defining meaningful metrics and interpreting results, including limitations and tradeoffs. - Ability to reproduce a baseline and provide reviewable research artifacts such as experiments, test fixtures, or corrections. - Clear communication of findings that can influence product or research decisions.

Benefits and work setup The role is part-time and may be based in New York City, the San Francisco Bay Area, or remotely in the U.S. or Canada. Periodic travel is expected for team gatherings and customer or industry engagements, including in New York, Washington, D.C., San Francisco, and customer locations.