CryptoVerif is a mechanized security protocol verifier sound in the computational model. In this paper, we explore and extend its treatment of dynamic key compromise. First, we present a basic treatment of compromise and explain its limitations. Next, we introduce several extensions in order to remove these limitations: improved proof of secrecy; building different proofs for the various properties to prove; removing code that cannot be executed when the adversary breaks the desired security properties; and guessing tested sessions, values, and branches. We illustrate how these extensions improve the treatment of key compromise on protocols ranging from toy examples to filling gaps in previous large case studies including TLS 1.3 and the WireGuard VPN protocol.