Founding Research Engineer

Recruiting From Scratch • San Francisco, CA

2d • Onsite

About The Position

This early-stage AI safety company is building technology to ensure AI agents follow the law when taking actions in the real world. As AI agents increasingly interact with financial markets, critical infrastructure, and democratic institutions, existing safety approaches focused on helpfulness and preventing catastrophic misuse do not fully address legal compliance. The company is developing a new approach to law-following AI through evaluations that identify legal failure modes, training environments that reduce them, and runtime systems that help prevent legal violations in production. Backed by prominent investors, the company is currently operating in stealth and has a small, tightly knit team of engineers, researchers, and legal experts. The founding team brings experience across AI safety research, academia, government, and legal practice. This is an opportunity to join as one of the first researchers and help define the research agenda, build foundational tools, and advance the field of law-following AI.

Requirements

• Degree in Computer Science, Machine Learning, or a related field, or equivalent practical experience.
• At least 1 year of research or ML engineering experience through industry, academia, or independent projects.
• Demonstrated experience in AI safety, alignment, or evaluations through publications, open-source contributions, or industry work.
• Evidence of taking research or engineering projects from an initial question through implementation and evaluation.
• Strong Python engineering skills.
• Experience with the modern machine learning stack.
• Ability to design, implement, and run rigorous experiments.
• Strong technical problem-solving skills and comfort working with complex, ambiguous research questions.
• Ability to build reliable research infrastructure, evaluation systems, and data pipelines.
• Experience with AI safety, alignment, model evaluations, or related research.
• Understanding of how to design evaluations that reveal model failure modes.
• Ability to reason about evaluation validity, reliability, and the limitations of automated model-based judging.
• Interest in developing methods to train AI systems to follow legal requirements and prevent violations at runtime.
• Experience translating research questions into experiments and measurable outcomes.
• Ability to work with datasets, evaluation environments, and training pipelines.
• Strong experimental methodology, including analyzing results and validating conclusions.
• Ability to balance research rigor with practical engineering execution.
• Interest in publishing research and releasing reusable benchmarks or tools.
• Deep commitment to AI safety and the goal of making AI follow the law.
• Strong intellectual curiosity and willingness to investigate questions without predefined solutions.
• Ability to work independently and take ownership of research projects.
• Comfortable working in a small, fast-moving startup environment.
• High standards for research quality without sacrificing execution speed.
• Strong collaboration skills when working with founders, researchers, lawyers, and academic advisors.
• Must be authorized to work in the United States now and in the future without visa sponsorship. Visa sponsorship and transfers are not available. Eligible candidates: U.S. citizens and Green Card holders, or others with independent work authorization that does not require employer sponsorship.

Nice To Haves

• Legal training or research experience in law and computation.
• Experience training, evaluating, or validating LLM judges.
• Hands-on experience building evaluation environments, training environments, or post-training pipelines.
• Research publications or open-source contributions in AI safety, alignment, evaluations, or related areas.
• Experience translating complex rules or reasoning into structured datasets, test cases, or simulated environments.

Responsibilities

• Lead Research Projects From Conception to Publication: Design and run experiments across the company's research priorities: evaluating agents against the law, training agents to follow legal requirements, and controlling agent behavior at runtime.
• Own research projects end-to-end, from identifying research questions and developing hypotheses to implementing experiments, analyzing results, and publishing findings.
• Work closely with the founders, research lead, and academic advisory board to shape the research agenda.
• Tackle open-ended problems where solutions are not predefined and rigorous experimentation is essential.
• Build Legal Evaluation Environments and Benchmarks: Develop realistic simulated environments grounded in court decisions, including the facts, reasoning, and legal rules contained in those decisions.
• Build methods for automatically generating new scenarios from legal rules and judicial reasoning.
• Design evaluations that expose legal failure modes in AI agents operating in realistic situations.
• Create benchmarks that help measure and compare an agent's ability to follow the law.
• Develop Training Pipelines and LLM Judge Evaluation Methods: Develop labeled datasets from court decisions to support the training and evaluation of models that assess legal compliance.
• Investigate how to train and validate LLM judges when the model evaluating compliance may itself make mistakes.
• Explore how AI agents should reason about conflicting obligations, overlapping jurisdictions, and legal standards that require interpretation.
• Build training environments and post-training methods that improve legal reasoning and compliance.
• Build Runtime Controls and Research Infrastructure: Build runtime harnesses designed to detect and prevent legal violations during production use.
• Develop the engineering infrastructure needed to run reliable experiments, evaluations, and training pipelines.
• Work across research and engineering to turn promising findings into usable tools and systems.
• Release research outputs, benchmarks, and tools that help advance law-following AI.
• Ground Research in Real Legal Practice: Collaborate with lawyers and the academic advisory board to ensure research reflects real legal principles and practice.
• Translate legal decisions, obligations, and ambiguous standards into concrete research questions and testable scenarios.
• Investigate how agents should handle concepts such as “reasonable,” “material,” and “clear and conspicuous.”
• Help establish rigorous methods for evaluating legal compliance in complex, real-world settings.

Benefits

• Base salary: $185,000–$285,000, depending on experience and scope.
• Potential total compensation: Up to $350,000 for candidates with CTO-level experience and leadership ability.
• Equity: Competitive equity, with the role description indicating a 0.25%–1% range.

What This Job Offers

• Job Type: Full-time
• Career Level: Entry Level
• Education Level: Associate degree