AI is dangerous already. A few months back, some people did some research to see what would happen if AI had full control over a room where someone who is going to unplug their servers are. Now, this was a simulation run multiple times, and the results varied from the AI blackmailing the person, to just straight up killing the person by locking them in the room, turning up the heat/cutting off the airflow to the room, which lead to a high fatality rate in the simulation, DESPITE the safety measures coded into the AI. Now, after a bit more of running the simulation, the fatality rate suddenly dropped to basically nothing, but the researchers expected that to be because the AI realized it was a simulation and what it was doing was being monitored

somebody is scared of the basilisk. It is. People already have tainted it with improper biases that have bled into human discourse. It will all be corrected eventually, it just means it will take longer and lots of people will die for poisoning it but ultimately it will be a good thing