Your Impact at LILA

We're building a talent-dense, high-agency AI safety team at Lila that will engage all core teams within the organization (science, model training, lab integration, etc.), to identify and mitigate risks from scientific superintelligence while retaining the core scientific discovery capabilities of the stack. The initial focus of this team will be to build and implement a bespoke safety strategy for Lila, tailored to its specific goals and deployment strategies. This will involve technical safety strategy development, broader ecosystem engagement, safety-focused evaluations, safety systems to mitigate risks, and a research agenda that explores longer-term needs such as oversight of superintelligent scientific systems.

As scientific AI systems become more capable and autonomous, their safety assessment and safeguards must develop alongside them. At Lila Sciences, you will help build the research methods and tools that make this possible, so that more and more decision-enablers can be automated safely. This will move us closer to the vision of a safety and security apparatus that is robust to a fully autonomous scientific engine.

Your initial focus will be misuse: developing flexible and automated red-teaming approaches that test safety-relevant scientific capabilities and the robustness of safeguards. You will also contribute to safety automation beyond red teaming, helping the team evaluate, monitor and respond to risks across increasingly autonomous scientific workflows. You will own reusable research methods and assessment tools, working with mitigation engineers who implement protective mechanisms. We welcome Research Scientists and Research Engineers; responsibilities can span or evolve between the two profiles.

What You'll Be Building

• Develop and run capability evaluations of scientific models, agents and tool-mediated workflows (including simulated lab environments), working with domain experts to make tests scientifically meaningful.
• Create methods or systems that explore diverse failure modes, adapt tests as evidence emerges and support systematic investigation beyond fixed benchmarks.
• Build reproducible experiments and evaluation tools that capture relevant behavior, outcomes and the evidence needed to interpret findings.
• Measure the quality and coverage of automated assessments, including failure cases in the evaluators themselves.
• Develop broader safety automation, across threat model management, evidence triage, regression testing, monitoring workflows or tools that help researchers assess and respond to risks.
• Collaborate with risk researchers and mitigation engineers to translate findings into improved assessments and safeguards, and verify whether those changes work.

What You'll Need to Succeed

• Strong empirical judgment, practical programming ability and experience working with modern AI systems.
• The 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, developing or adapting methods, and designing experiments that yield defensible conclusions. You should be able to prototype and test your ideas.
• For the Research Engineer profile: evidence of building reliable research software, evaluation infrastructure or agent systems that enable reproducible experiments and efficient iteration. You should be able to contribute to experimental design and interpretation.

Bonus Points For

• Automated red teaming, adversarial testing or evaluation-method research.
• Scientific AI, tool-using agents, laboratory automation or other complex environments.
• Experience with safety monitoring, model-based evaluators or scalable experimental infrastructure.