Am I the only one who thinks about 50% of these are played up for PR? Like sure, the first few might have been real, especially the one where they hit huggingface, because huggingface actually had to respond and stuff, but I'd be willing to say this is a "fable" situation, where danger is played up for dramatic effect, making them appear more impressive and adding to the mystique.

Like sure, AIs are dangerous, but not really in those ways. More in the ways of information and trust, our democratic process, copyright and ownership of ideas, social engineering, impersonation etc.. They do find exploits a good bit but it's hard for me understand how multiple companies accidentally had broken sandboxes and then in the span of 1-2months we have more than 4 reports of the same thing happening in the same way to multiple companies. These odds are worthy of distrust.

Tried

So it spat out some text that said "I'm now connecting to ChatGPT beep boop beep beep that's me dialing for realsies" and now we get to pretend something newsworthy has occured