Nope, you misinterpreted. It generates truthful messages from truthful intents. In order to do this, they do imitation learning from human dataset, filtered by lie detector. They filtered 5% of human turns for lying by human.

Edit: This is why they had to create a specific filter to avoid confessing plans to stab, because otherwise it would confess. The resulting system still does not lie to the player it is about to stab, although it may remain silent or talk about other things.