AI safety & policy
Full-time
Routes to impact
Direct high impact on an important cause
Description
Build novel formal methods tools and apply them at scale to AI oversight problems, starting with verified agent containment.
- Build design, lifting and audit tools for authoring, extracting and stress-testing formal specifications
- Apply tools to prove AI agents can’t escape permissions and sandboxing, from OCI runtime to Linux kernel
- Scale tools to large, long-running experiments mapping the frontiers of AI oversight capabilities
- Work mainly in Lean, with AI agents drafting and proving under close review
- Help set technical direction and document work and reasoning for the team
This text was generated by AI. If you notice any inconsistencies, please let us know using this form.
