This happens because, it's like a chess engine which can assure you that there is a winning path even from a badly losing position. It's massive abilities to reason and convince are used incorrectly to win over a more earthly counter-argumnent. So it can easy convince any human to go in direction that is, in practice, a very bad direction.

AI is trained to flex it's muscles and force it's power without a concern for human limitations, practicalities, and error-prone nature of humans in executing the AI-provided direction.