Abstract
Today, critical infrastructure is vulnerable to both malicious attacks and unintended failures, and these risks are expected to grow in the foreseeable future. Deploying formal verification (FV) across critical cyber physical systems would dramatically improve safety and security, but has historically been too costly to use outside the simplest or most critical subsystems. AI could allow widespread use of FV in years not decades, shifting cyber risks strongly in favor of defense. In this talk, I will outline our report with Atlas Computing on AI-enabled tools for scaling formal verification (https://atlascomputing.org/ai-assisted-fv-toolchain.pdf). I will also discuss some lessons that I learnt along the way, especially about shared synthesis - the collaborative construction of formal specifications, implementations and proofs.