They're definitely developing a new model on vetted public domain data as we speak. They just need to delay legal action long enough to get that new model to launch.
This is the same thing YouTube did. Delay all copyright claims in court, blaming users, then put their copyright claim system in place that massively advantages IP owners.
Who knows how the laws will change because of AI. But as the law currently stands it's just a matter of proving it to a court. That's the main barrier.
This is strong evidence an AI is breaking the law.
The liability of industrial machines is actually quite apt.
If you design a machine that kills someone during reasonable use. You are liable.
Aircraft engineers have a 25 year liability on their work. A mistake they might make could kill hundreds.
There is always a human responsible for the actions of a machine. Even unintended results have liability.
If you upload a program to a machine and someone dies as a result you're in hot water.
Moving away from life and death, unintended copyright infringement by a machine hasn't been tested. But it's likely it will be ruled that at least some of the builders of that machine are responsible.
AI "self-driving" cars are getting away with it by only offering an assist to driving. Keeping the driver responsible. But that's possible because you need a license to drive a car in the first place.
AI images like this are the equivalent of a fully self driving car. You set the destination, it drives you there. The liability falls on the process of driving, or the process of creating. The machine doing that means designers are then liable.
Except AI models may end up having to start again with licences or public domain data.
They are currently breaking the law and delaying legal action as long as possible in the hopes they can repeat the trick with a new data set.