I don't think it can lie.

Seems like it provides the most predictable response based on the context and training.

Now, an interesting question is: are llms a different kind of mind that don't experience truth like we do?