Astrio is an independent research group advancing AI safety through formal methods.
We work on one question. When a machine produces an answer or takes an action, what does it take to know that it is correct rather than merely convincing?
We build and study systems that make AI more verifiable, from formal verification and proof assistants to neurosymbolic systems, automated reasoning, and agents that operate in formally specified environments.
Our goal is to develop methods and infrastructure that give stronger guarantees about increasingly capable AI systems.
Everything we build is open.
