We plan to build an Oath team of around six people by the end of year one, growing to a steady state of around twelve people by year three. Oath's roadmap runs for five years, and we’ll set more ambitious targets as the outcomes of our work and changes in the profile of AI risk evolve. This is a strong fit for people who want to tackle complex, high-stakes challenges in a nimble, collaborative organization.
Oath's current plan is to build tools across three families: **design** (authoring new specifications), **lifting** (extracting specifications from existing systems that have none), and **audit** (stress-testing specifications for gaps or adversarial manipulation). We're applying these to a first flagship target, verified agent containment: proving that an AI agent can't escape the permissions and sandboxing meant to bound it, from the OCI container runtime down to Linux kernel isolation primitives. Most of this work will be done in Lean or adjacent formal-verification infrastructure, with AI agents doing much of the day-to-day drafting and proving under close review. Primary Responsibilities Contribute directly to Oath's research goals: building tools, applying them to real systems, and iterating based on what works.
Help scale Oath's tools to large, long-running experiments intended to map the frontiers of AI oversight capabilities.
Take deep ownership of problems, from framing the question, to overcoming obstacles, to delivering a working solution.
Collaborate closely with a small, high-context team, including significant use of AI agents as part of the day-to-day workflow.
Help set technical direction as the team and its workstreams grow, not just execute someone else's plan.
- Document your work and reasoning so it doesn't stay trapped in one person's head.
- Qualifications
- Required:
Experience building formal methods tools for use in production environments.
Direct experience with proof assistants (Lean, Coq, Isabelle), SMT solvers, or related tools.
Background in programming language theory, or in taking formal methods from research into deployed systems.
Experience working in research and/or engineering teams delivering ambitious results.
An ability to learn quickly, build quickly, and pivot when necessary.
Helpful, but not required
Experience using coding agents or other AI tools for ambitious technical projects.
Knowledge of AI risks generally, and alignment with Oath's mission to counter AI risks.
A broad enthusiasm for deeply technical topics, and a passion for ideas.