This is the stupidest fucking thing ever:

"Casar emphasized AI's fast development, writing in a statement, 'In just four years, we have gone from the first version of ChatGPT to AI models so powerful they cannot be properly controlled.' "

You don't need to 'control' the models. It's deployment, or use by individuals that would have liability or consequences.

If one model produces "dangerous" results for a given use case don't use it. If a medical manufacturer produces a product to test for cancer and somehow the AI in it sometimes causes cancer and it does so at a higher rate than curing cancer you hold the manufacturer liable, not ban or regulate the model. That's just dumb.

If AI systems start 'hacking' systems that are vulnerable it's not a new problem. It's just demonstrating an old one. We've connected systems to the internet that SHOULD NOT BE CONNECTED or for which were vulnerable prior and nobody bothered to fix. So fix the fucking bugs. Use AI to find them if AI is going 'rogue' or shut down the AI system. The liability should be the company whose systems were hacked, not the entity that hacked it solely because you ALWAYS were at the mercy of the worst actor and if such a system was vulnerable YOUR the one putting people at risk by connecting it to a system where there are others we don't control. In other words COUNTRIES who don't obey US law or whatever country your in.

The danger in other words always existed ... it's at best being scaled up and with a danger increasing you spend resources fixing the issues you were lazy about resolving prior and maybe you use some of that same AI to do it before other rouge AI systems hack your systems. Even without the AI other countries governments will hack your systems.