The "AI" most people have been exposed to are very good word prediction machines that produce mostly wrong answers only. It's quite stupid, but the piles of random matrices that produce recognizable text have appeared as magic to the world.
The danger is not the "Skynet" or sci-fi takeover cliche, it's the fact that so many people put their faith in the bullshit machine. Leaders can just say "this new medicine uses the [bullshit machine] to be made perfect" and people who were enamored at ChatGPT4-o1-extended-Claud-lama being able to write an original Sci-Fi story about Rumpelstiltskin on an alien planet filled with sentient insects and cowboys will see the claims of perfect AI generated medicine as absolutely safe and effective.
The bullshit machine is not the danger. The belief in the bullshit machine is the real danger.