Comment by jurgenburgen

> Hacking incidents this summer show that big labs struggle to control models, while models are increasingly aware of the training process.

I think that’s more a sign of their priorities. Controlling AI doesn’t increase value, so there is no incentive.