Software is used across every field today, and the errors and security vulnerabilities inside it can cause serious incidents and damage. The Information Security Laboratory studies how to detect such errors automatically and remove them in advance. It uses static analysis, which approximates a program's behaviour and predicts errors without running it, and fuzz testing, which finds errors by running a program repeatedly on varied inputs, to detect faults in software from many domains. Recently it has been combining the strong code understanding of large language models with program analysis and testing to open new possibilities in software error detection.