Analysis found 434 exploitable flaws in AI-generated apps, with denial-of-service, authorization and secrets exposure risks among the most common issues.