In a liability environment, you attempt to describe the true cost and risk and allow the company to adapt to changing environments.
A practical real-world difference is "password rotation requirements". Most real-world security professionals knew the dangers of strict password rotation requirements for years before NIST could release information on it. And just because the process of standardization must necessarily be slow, the NIST requirements would then have to flow to other departments so they can then update their standards and so on.
Today, in most financial and healthcare companies, password rotation standards are rampant despite it being the case that NIST has advised against. This is often because the companies don't want to spend the money to alter policies but also often because in the 'no-man-is-an-island' interconnectedness of firms, policies in one can impose corresponding policies in others. So that means that your new startup will need to rotate passwords every month in order to integrate with (say) Bank of America.
Apart from all that, if you genuinely think about it, we want to do cost-benefit analyses on this. If the risk of leakage of customer data is low enough and we value it at some $x dollars per unit, then there's some $y of cost above which it isn't worth it. This intuitively makes sense since customer data, no matter how personal, isn't worth infinity. If it were, no one would collect it. No one. In fact, by giving me your phone number I would suddenly be holding an artefact of infinite value. Or by giving Amazon your shipping address. No one wants that liability and information exchange would halt despite everyone (in reality) wanting it to happen.
What I hoped for is that I, as a startup, can beat the government on picking a security standard because the government has to cater to all but I can just beat it by being better than Big Slow Dinosaur (BSD). But because I have to integrate at some level with BSD that has to follow the government I lose that advantage.
I suppose, with a regulator capable of moving fast to respond to threats, aware of sunset periods, regulation could be superior. Somehow, I expect forums like this one to be full of software engineers complaining about "the constantly changing requirements from NIST" if that were to happen.
A Liability regime would encourage companies to be actually secure, because they're responsible for what happens to data that is lost. A Control regime would encourage companies to check boxes from a list provided by a regulator. A Liability regime encourages being pro-active vs reactive to the regulator in a Control regime.
There are middle grounds, like HIPAA or GDPR. Both give companies some leeway in terms of creating their own checkboxes, and fines are for actual breaches, not just improper process.
In a regulatory environment, if a corporation does not comply, the punishment is increased until either the corporation complies or seizes to exist. Since not existing is bad for profit, corporations usually comply eventually.
In a liablatory environment, as long as nothing happens, a corporation can continue to proceed down a dark path with no ill effects and abuse the rules as they see fit. It's only costly when things go wrong and you can calculate the likely cost of things going wrong.