2 articles on hallucinations.
The dangerous AI answer is fluent, specific, plausible and wrong in a way you will not discover for three weeks. Detecting those is mostly mechanical. Here is the mechanism.
AI doesn't 'hallucinate' randomly — it invents code when it can't see the system. Give it structure and validation, and the hallucinations stop being your problem.
