Ryan Bridge: Is our data safe with AI?

The more you read about these AI hacks, the less you want to know about them. 

Basically, the short story is that there was an AI bot out there that went full Terminator. It hacked into another system, another AI system, without being told to do so, acting entirely of its own volition. 

It was reportedly an OpenAI model that did this. It escaped the environment where it was supposed to be, went online, and stole the login details for Hugging Face, which is the other platform, right? 

The interesting thing is what Hugging Face, the victim in this case, did about it. They said that, normally, you would get another AI bot to attack the attacking AI bot, and they would have some kind of war. But they turned to a Chinese bot to do it. 

Why? 

Because the American bot had so many guardrails on it that it had essentially rendered itself useless. It wasn't good enough to defend their system, so they went with a Chinese one. 

Fascinating. 

So, you basically have robots at war with other robots with no real control over what they might do next. And we've seen that before. This is a clear-cut example of it, and it's not the first one. 

OpenAI has said they've had issues before. Apparently, they've previously admitted that their models autonomously took unwanted actions for extended periods to achieve their goals, including looking for ways to act outside their sandbox. 

Well, that doesn't sound very good for us, does it? 

My question is this: if you have robots attacking robots, Chinese robots attacking American robots, and robots doing whatever they want, how safe is any of our information every time we log into a platform? 

How safe is that information? And, therefore, what information are we actually willing to give to those platforms? 

See omnystudio.com/listener for privacy information.