AI Safety Research Scientist — Red-Teaming & Evaluation

Lila Sciences, Inc.; Cambridge, San Francisco (MA, CA); On-site; USD 180,000 - 298,000; Full time

Job summary

Lila Sciences is assembling a Safety AI team in Cambridge, MA, London, and San Francisco to develop tests and tools that evaluate safety-relevant capabilities of scientific AI models and workflows. You will work on automated red-teaming, safeguards, and evaluation infrastructure to advance responsible autonomy in scientific AI systems.

The role spans Research Scientist and Research Engineer profiles, focusing on empirical methods, reproducible experiments, and collaboration with domain experts

Qualifications

• Strong empirical judgment, practical programming ability, and experience with modern AI systems.
• Ability to investigate ambiguous failures, design informative comparisons and communicate limitations clearly.
• An interest in scientific misuse risks and the ability to collaborate productively with domain specialists.
• For the Research Scientist profile: evidence of selecting valuable research questions and prototyping ideas.
• For the Research Engineer profile: evidence of building reliable research software and evaluation infrastructure.

Responsibilities

• Develop and run capability evaluations of scientific models, agents and tool-mediated workflows.
• Create methods or systems that explore diverse failure modes and adapt tests as evidence emerges.
• Build reproducible experiments and evaluation tools with evidence capturing outcomes.
• Measure the quality and coverage of automated assessments, including failure cases in evaluators.
• Develop broader safety automation across threat model management and monitoring workflows.
• Collaborate with risk researchers and mitigation engineers to translate findings into safeguards.

Skills

Empirical judgment

Programming

AI systems

Experiment design

Communication

Collaboration

Job description

Lila Sciences is assembling a Safety AI team in Cambridge, MA, London, and San Francisco to develop tests and tools that evaluate safety-relevant capabilities of scientific AI models and workflows. You will work on automated red-teaming, safeguards, and evaluation infrastructure to advance responsible autonomy in scientific AI systems.

The role spans Research Scientist and Research Engineer profiles, focusing on empirical methods, reproducible experiments, and collaboration with domain experts