AIUC Labs Research

Artificial Intelligence Underwriting Company•San Francisco, CA

About The Position

AIUC builds standards and insurance for frontier AI. Our mission is to Underwrite superintelligence. We exist because risk, not capability, is becoming the binding constraint on AI being useful. History offers lessons for how to build confidence in new technology; electricity burned down houses until insurers funded Underwriters Laboratories (UL) to certify products against their standards. Today, the UL mark is a trustmark on every light bulb in America. We’re building that for AI. AIUC-1, our standard for agent security, serves frontier builders like Cursor, Lovable, ElevenLabs, and Harvey. We've raised $55m from Ribbit, First Harmonic and Nat Friedman to audit frontier AI. Our team comes from METR, Anthropic, McKinsey, and OpenAI. Come join us. The Role Superintelligence raises imminent questions: What models are safe to be released? How is the progress of recursive self-improvement audited? How should lab incentives be shaped? Who governs self-sovereign agents and what should incident forensics look like? Fortunately, researchers, auditors, hackers and actuaries have grappled with older versions of these questions for decades across nuclear energy, electricity, and car safety. AIUC applies that same combination to AI. AIUC-1 lets us underwrite the world's first agent insurance policy. We hold eval data across the most-used AI agents, a team with both frontier AI and traditional risk experience, and a consortium of Fortune 1000 security leaders. Answering these questions in time means combining frontier AI research with risk tools that already work. That's why we're founding AIUC Labs – a research lab which will exist so that AIUC and the world at large is on the frontier in navigating AGI. You will work directly with the founders to build out a new function and publish research that will guide the industry on what’s next.

Requirements

You’re a strong engineer, and can design and independently build, debug, and maintain research code and experiment pipelines.
You can read an ML paper and implement it quickly, and extend or adapt it to a new research question.
You can iterate and experiment quickly, such as working through an ambiguous open-ended question and turning into a tractable experiment, choosing baselines, and investigating unexpected details, as well as deciding what comes next.
You have a strong background in ML, CS, or an adjacent quantitative field, developed through a combination of engineering and research.

Nice To Haves

You might be coming from a frontier lab, a third-party evaluation organisation, or an ML team at a technology company in e.g., science of evals, security research, forecasting, or interpretability.
You’re familiar with the model training pipeline, and have trained, fine-tuned, probed models, or built infrastructure that makes experimentation scalable.
Published work that people in the field actually engaged with: a first-author paper, a benchmark others adopted, shifted how a community thought about something, was built into a product, or changing how real systems get deployed
Range across the philosopher-to-engineer spectrum — comfortable arguing about where capability is heading in three years, and comfortable writing the code that tests whether you're right.

Responsibilities

Own the questions that decide the strategic direction of AIUC: where AI capability & risk goes over the next 3-5 years, what replaces evals as agents get more capable, whether models behave differently when they know they are being tested, and how AI shifts the offense–defense balance in security
Take confidence infrastructure upstream to the model layer, where independent standards, rigorous evaluation and third-party assurance have to meet the release decisions labs are making now
Turn what AIUC has actually done into evidence, publish research and shape our public research programme, including the podcast and writing, that increases our influence in the space