A proof assistant is a software system in which mathematical theorems and proofs are written in a formal logical language and the computer mechanically checks that each inference step follows from the axioms and rules. Such systems are also called interactive theorem provers. Users typically write the structure of a proof and fill in routine steps with automated tactics; a proof that passes the checker is guaranteed to have no logical gaps. Prominent proof assistants include Lean, Coq (now Rocq), Isabelle, and Agda.
Proof assistants were originally used mainly for the formal verification of software and hardware, but they have become increasingly important in mathematics for checking large proofs, with Lean's Mathlib library at the center of this activity. In AI, proof assistants are widely used to train and evaluate mathematical reasoning models, since a proof generated by a language model can be checked automatically to determine whether it is correct.