> It will depend on the consent of the Beneficiary whether to turn back the Funds to the Originator.
SCT Inst Scheme Rulebook § 4.3.2.3 [0].
Of course, if you are talking about fraud, yes: banks may answer positively to a recall, without contacting the beneficiary, when the beneficiary is a fraudster. Indeed, local laws often require that of banks, and therefore the standards define a dedicated protocol for that.
When you break the law, many legal protections no longer apply.
There are many cases where asking the beneficiary is not necessary, fraud is only one of them.
And to clarify, it's not really "fraud", it's "claim by the originator that a fraud occurred". Which is a very different thing, and which matters immensely in practice.
There are three cases. Fraud is one, the two others are banking mistakes. First, when the bank sends the funds twice by accident. Second, when the bank was not meant to send the funds.
In both of those cases, the beneficiary bank is not even required to send the funds back. It can require consent from the beneficiary (which is its client, after all) if it wants.
Yet in both cases, the beneficiary is not even meant to receive the funds.
Fraud works the same way: unless local law requires it, the beneficiary bank is not required to send the funds back, and can rely on beneficiary consent.
> 2 more waiting to be identified among the running product
I don't know anything about this factoring project; does this mean that keys are somehow "tested" in batches, and if the batch comes out "dirty" then you know you can easily factor something in that batch (for some value of "easy")?
EDIT: Ah, nevermind, just figured it out. For those coming after, there's a running product of all public moduli in the tested set; you can test one key against all of them simultaneously. When you find a GCD /= 1, you've found a factor, and you know the key you just tested has it, but there's an older key which has that factor too, and you don't know which one. Presumably you either go and test them one by one or store "snapshots" of the running product to test.