The short answer is, yes, they are for now, in the sense that we don't think current models are capable of doing the supergalaxy-brained, long-term biding of time to do some big takeover. However, that is the plan—making them that smart. I think we're probably within six or 18 months of models that can do this, though I'm not saying that it's going to happen in that time frame.
This was a big crisis of faith within the company. They had this responsible scaling policy, RSP, which I helped work on a bit. The idea was that they would never ship a model they couldn't prove was safe. However, they realized that they had backed themselves into a corner. They couldn't prove the models were dangerous, but they couldn't prove they were safe. If they unilaterally stopped, they would blow up the company, to no one's benefit. They just changed that RSP two weeks ago to remove that provision.
The point is that they know they're entering a regime where they can't prove the models are safe anymore, but they also don't have a great plan for dealing with that. They wish everyone could slow down, but that requires coordinated action.
