A chatbot generates fluent text that contradicts the company's published policy. Which explanation is most accurate?

1. A longer answer is always more accurate
2. Fluent generation is not proof of factual correctness
3. Grammatically correct output is guaranteed to match every private policy
4. More sampling randomness guarantees policy compliance