Core adding full RBF is a change of node policy that may be highly inconvenient for zero-conf users, but there has always been and will always be a risk of a double-spend for anyone that treats zero-confirmation transactions as settled. It's literally in the name - this transaction has zero confirmations and no guarantee it'll make it into a block, and so has not yet settled.

The perception seems to be that Core adding the full RBF option is increasing the risk to zero-conf users, but I'm not convinced that that is the case - someone wanting to double-spend attack you isn't going to be bothered to do so over a few thousand sats (unless they can do it thousands of times), and losing a few thousand sats to a double-spend isn't the biggest deal.

It's always been the risk of getting double-spent out of hundreds or thousands of bitcoins that's worth seriously worrying about, which is much more the kind of attack a determined attacker is able to carry out. Such a determined attacker is much more likely to attempt and succeed at a sybil attack, or directly colluding with a miner. So your zero-conf risk increases non-linearly as the amount of bitcoin being transacted grows. (caveat: this paragraph is opinion).

There does, however, seem to be a legitimate business for providing insurance/risk management for people that are willing to accept the zero-conf risk - it is pretty similar to accepting credit cards with a chargeback risk or any payment card with a capture risk, though there's no-one to mediate a dispute. On-chain is final.

But what doesn't make any sense is trying to avoid Bitcoin Core and nodes from adopting a full RBF policy to try to protect this use case. As has been pointed out by may others before, full RBF is aligned with miner (and user) economic incentives and is a node policy, not consensus, so you can't even tell which nodes are doing it nor can you prevent them from doing so. Second, Bitcoin core 24 with the full RBF option is already out in the wild at around 5%+ of running nodes and growing, so it's too late to kill it.

So my point is that relying on node policy as part of your protection for zero-conf transaction acceptance is fragile, and should not be relied upon. The protocol rules have always tacitly allowed double-spending before a confirmation, and it has always been clear that there's no consensus on which transactions have occurred until they have in a block and have at-least one confirmation.

The long-term 'what to do about it' is to use Lightning if you want fast payments with risk-free instant settlement, or as above, accept the zero-conf risk and cover yourself with an insurance premium (e.g. a margin on transactions that goes into an insurance fund, and limiting max transaction amount so you're not exposed to uncoverable losses if you do get double-spend attacked)

Angus