I imagine it's possible to design a blockchain where transactions are 1) artificially delayed; and 2) can be blocked by anyone who has the key.

For example, take Bitcoin, except that each transaction must be posted twice to be considered to have modified the ledger between 7 and 14 days apart as measured by the number of intermediate blocks as a function of the difficulty requirement. If a "blocking" transaction makes it to the blockchain in between, then the transaction is deemed to have been cancelled. A "blocking" transaction must merely demonstrate ownership of one of the transaction inputs.

If you hold large funds, you could keep a remote monitoring system somewhere that is empowered to block your own transactions.

The blockchain design could be enhanced so that a blocking transaction proves ownership of the original inputs but cannot sign a transaction itself.

Failure mode: someone trying to legitimately block a transaction could find themselves unable to get their blocking transaction mined. But in practice they'd just up the transaction fee, and game theory should take care of it.

This was just a thought experiment. A point that comes to mind is that given this is possible, what you describe isn't an inherent property of cryptocurrency, so if there's a benefit to having this property then it could be done.

It wouldn't work if the realisation of fraud takes longer than the window of time you've given the owner to make a blocking transaction. Banks are superior here because they have the power and ability to correlate multiple different kinds of evidence, and then reverse the transaction even years later. They have financial statements, biological proof, social evidence, much more than just one cryptographic key.