At Oath, we’re building tools for oversight of advanced AIs. As AIs become more powerful, it will become more difficult to understand and monitor their behavior. Oath is building oversight tools based on formal verification: rather than monitor AIs directly, humans define unambiguous rules, and AIs must provide trustworthy mathematical evidence of compliance. In this way, Oath will help keep humans safe and in control as AI advances. Our work combines specification design, formal semantics, verification, and auditing, with an initial focus on verified containment of AI agents and applications in critical software and infrastructure.