That's certainly one reason, but frontier LLMs are also really good at finding existing vulnerabilities in completely human-authored codebases.