"I think the big difference is what we define as cheating. Optimising towards test, no cheating in my POV. Using one type of steel, as an example, for crash test and something different for production -> cheating."

Is it still cheating if there is a loophole that allows that substitution of steel?

Very interessting edge case. Legally no. Personnaly I would say no as well, morally it would be absolutely unexceptable. But then the loop-hole is the issue and not the use of it, right? Same goes for the Toyota example montioned above. The optimized for an existing test, fair enough. Testing changed, so Toyota has to adopt (if they didn't already). Again not really cheating.

I think that normal people internalize a multi-layered context of ethical rules. Following the letter of the law is only one layer. Hence most people would not say "fair enough" just because regulations were followed, if the meta-rules of society were not. I don't think it is possible to do away with meta-rules by writing more detailed laws, although I get the impression that is an increasingly popular assumption. In my opinion, the ubiquity of computers leads to an overestimation of our capability for directly handling complexity in rule-based systems.