Yes, it's complete hyperbole, on multiple levels.

"AI" like what people are hyping now, isn't telling anyone to do anything, on account of it being merely a statistical summary of a large set of texts or images. It doesn't understand anything in any meaningful way. It doesn't think. It doesn't have opinions. It's not aware. And the systems we have now fundamentally do not scale to that.

What they can do is you give them some input, and then their internal mathematical model provides you with some output that statistically seems to relate to what you provided. Maybe the response is grammatically correct - the system doesn't know, or even haven the capability to be able to know. Maybe it's semantically valid - the system doesn't know or have the capability to ever know. The system also can't know if the response is true or false or gibberish - only that there's some statistical correlation to the words it has been fed earlier.

They also don't learn or remember what you or they said. In order to appear to be engaging a conversation, they are not given an input of what you just wrote, but an input that also contains all the previous things you wrote in that "conversation" and what the system replied back. Then it gives you something that statistically relates to all that input.

If you initially give them a lot of text like "KILL YOURSELF MUH CLIMATE", they are likely to repeat some of that back. If you give them the bible, they are likely to repeat some of that back. As mindlessly as your calculator computing 2+3 for you.

There are certainly entities out there that could be said to be satanic, and want you to off yourself, for the climate or for any excuse they can sell you. But those are not overhyped statistical text models called "AI".