Ultimately all this stuff comes down to risk and risk management. The card networks themselves don’t provide any technical enforcement of most of their rules, beyond low level technical rules around message format and structure (and even that enforcement is pretty small).
Instead they effectively make all the parties that connect to the network responsible for rule enforcement. If a merchant follows all the rules correctly then they receive extremely strong chargeback protection, I.e. if an issuer sends a chargeback, and the merchant has plenty of grounds to dispute the chargeback and win.
If merchants don’t follow all the rules, then issuers can send chargebacks, and it’s much harder for the merchants to defend themselves.
In all scenarios it up to the issuers and merchants to explain in the chargeback what rules have been broken, by which party, and thus who should win based on network transaction rules. The networks themselves don’t even make a ruling directly, instead the issuer and merchant decide who wins via a back and forth process that includes escalating fees paid to each other, until one side gives up. The networks only get involved the two sides can’t resolve the issue themselves, and will charge the looser a significant fee for the privilege, so there’s a strong incentive for the parties to resolve the issue themselves.
How does of this interact with 2FA, auths etc etc. Basically 2FA, and ordinary auths are all just things a merchant can do, or trigger, to reduce their liability and get better chargeback protection. If the merchant performed a full 3DS auth, where the issuer is asked to perform 2FA, then they have pretty complete chargeback protection in the event of fraud, because they’ve basically asked issuer to make absolutely 100% that this transaction has been approved by the issuer’s customer, so there’s zero grounds later to claim that a stolen card was used or something similar. If the issuer’s customer wants to dispute the transaction, that’s the issuers problem.
But all of these mechanisms reduce checkout rates, and thus merchant revenue. As a result some extremely large merchants make a trade off of basically accepting all the risk of fraudulent transactions, and give up chargeback protection, but not following all the rules. The merchant does this because they’ve basically asked believe they have enough data to prevent fraudulent transactions, without using any of the tools the card networks provide.
For merchants that can do this (like Amazon), they build in-house fraud detection systems, and payment systems that evaluate the risk of each transaction, then change the exact way they perform the transaction to either reduce friction (because the transaction is very low risk) or increase friction (because the transaction is higher risk), thus allowing them to capture more revenue, without taking on more risk (because they have confidence their ability to detect fraud, and thus don’t need help from the issuers).
But there are very few merchants that can even do this, as it generally requires either a very collaborative payment gateway (who are ultimately on the hook for merchant misbehaviour), or a direct connection to the card networks (who aren’t interested in talking to people not moving millions of dollars every day). Which is why it tends to pretty rare.