That's not the way OpSec works guy.
I realize my view is strongly informed by the military... but I view this as essentially OpSec for the civilian leadership, and the principles can't be that different. I may be wrong in my interpretation of this as OpSec, but I doubt it.
If you need that clause in an agreement, FOR ANY REASON... then you need to re-evaluate the agreement. Full stop.
Cooperating government Z gets email Y, and discovers it's from senator X because they look more closely and connect dots the US may not have. No process could have prevented this (except stopping the sharing entirely). It's a reasonable clause.
Of course that wouldn't scale to a system that shares a big firehose of data or does that in real-time. Easy to solve too: don't share the firehose, and don't share in real-time. Collaboration with allies is fine but that should come as a distant second to protecting your own [citizen's] interests.
A manual process could certainly do a better job, but no, it still isn't a guaranteed thing.
What makes you think the NSA is sharing data with Israel for any other reason? We have strategic and human interests where they are located.
So the only secure path seems clear.
To me at least.
Again, maybe this information is not as critical as information you might be exchanging in the military. I don't know. But, to me, it seems a good example of information you don't want foreign powers being privvy to.
My gut would tell me that you want as perfect OpSec as you can manage, but then have contingencies for the occasion when something fails. Otherwise you would get the kind of flailing we're seeing right now after the Snowden leaks.
1) When classified information is accidentally moved to a computer with a lower classification it is a shit storm. Everything stops. All hardware involved is isolated and scrubbed. If it can't be scrubbed with certainty (e.g. deallocated sectors on the hard disk that might contain classified data but can't be scrubbed anymore because they are deallocated) then the equipment is destroyed (or classified to match the level of the information that was leaked).
2) All information on a classified system is considered classified at the level of the computer system it is stored on. When there is a need to move not-really-classified information off that system (like send a log file that a vendor needs to help debug software) then it is manually inspected by a SME (subject matter expert) to positively verify it does not contain classified information and then copied off to unclassified media. But if there is any doubt, like a binary core file, the file is forbidden from leaving the system.
The main point here is that the SME positively identifies all the data as being unclassified. The assumption is that the data is classified and it must be pain-stakingly verified on a line-by-line basis. It is a real PITA.
Considering that these NSA programs are highly classified, at least TS and I'm thinking TS sci/codeword given the style of single-word names for the programs, the same sort of manual, positive identification process should be used when giving the data to another country.
Also, Israel is considered a major intelligence adversary by the US DoD. Israel and China are at the top of the list of countries they warn people with clearances to be careful of. Blithely handing over this raw SIGINT data is ridiculous given that context.