There are a whole lot of combinations that don’t work out well for the interlocutor.

For example, if “is it safe to be near the pangolin?” is anything above neutral and “can diseases spread through contact” are either thumbs up or double plus thumbs up, then it doesn’t matter how emphatically health guidelines tell you otherwise, this AI wants you to lick that poor thing and roll the evolutionary dice.

So maybe it is an AGI?