Someone could turn your system against you

Even after training is done, the model operates on hardware that can be updated. Which means if someone hacks it, they can update the AI to be or do whatever they want. Which is incredibly dangerous.