All of them will leak? Seems like if each key is in possession of a different party, and you need all keys to decrypt, and at least a few are air-gapped only available for warrant holders via manual retrieval... the probability would be vanishingly small.
I agree it's impossible in theory, but in reality nobody has put out significant resources and made an honest effort to try, imo.
I bet I could come up with an encryption scheme that allows decryption for warrant holders only, but the scheme to decrypt will be very onerous for the warrant holder and will require manual retrieval of air-gapped asymmetric keys and lengthy background checks to establish trust (i.e. is this person who they say they are and did the judge really grant a warrant, etc.).
There are always tons of assumptions we make when we say these things.
We assume that the computer used to generate the keys was never connected to the internet, has not already been compromised at the time of key (re)generation, there are no backups of the computer, the engineer that performs the key generation is extremely trustworthy, the room this is done in is perfectly isolated from surveillance, etc. If you start calculating these probabilities and multiply them together, you start observing them moving further and further from 100%.
> but in reality nobody has put out significant resources and made an honest effort to try
These just seem like No True Scotsman fallacies. If anyone comes up with evidence of an effort, are you going to move your goalposts?
IMHO, the incentives don't exist. Law enforcement can't even secure their own technology and vendors (eg. BlueLeaks, OPM hack). The tech companies tasked with this project only see it as a cost center and see it as a potential massive risk to their brand if this system is misused. During the Snowden leaks, there were accusations of NSA contractors using the search system to spy on family members, dating partners, etc.
Human behavior is too predictable to naïvely assume that we can increase security while increasing the number of keys/people who have access to a massive tranche of data.
The nature of the federal executive is a problem here: keys spread among federal agencies are functionally in the hands of one party.
But I agree, it's a perfect example of why we can't trust even the most security-trusted people in the US government with skeleton keys to all of our data.
You can add whatever extra bureaucratic processes you want, but you're still making my encrypted data less secure.
You say that like there isn't form. The keys used by every DRM scheme has leaked. Every one. It was not from lack of trying, not because they didn't have access to the best and the brightest. HDMI was actually pretty good.
The problem is not just technology. It's also humans. Create a big enough prize for obtaining a single thing, and it will leak. Every man has his price.
When the prize is the ability to read every communication of every USA citizen, politician and corporation, if you aren't thinking about what sort resources nations like China and Russia can throw it at, you aren't thinking hard enough.
Whilst in theory ideas like these work, they don't in practice.
When security is tightened, and oversight is made more difficult, we regularly see the players involved skirting the system because they're human and don't think that they're really doing anything wrong.
Stuxnet was a while ago, but pierced an air-gapped network through a means that... Shouldn't have been possible. It was possible because of the failure of the humans involved. A factor that can't be removed.
You would find keys being surreptitiously shared over insecure shares, and getting caught up in other data leaks. Two or more departments sharing keys between employees to save some time and effort. Or forgetting to shred a key securely once it has been accessed. It doesn't take long until everyone in the department ends up with a key - and then each of those is a weak point.
The NSA has had repeated problems with their processes being avoided and ignored by employees skirting oversight to stalk their ex's, by doing things like listening to their phone calls - which supposedly required a warrant. [0]
You can't trust the people involved.
Now, there's a very, very simple way to avoid all of these problems. Secure the data to the fewest number of people required. Placing the responsibility of keeping the keys secure with the direct stakeholders.
An end-to-end encrypted chat still has the possibility of keys being leaked. But only the people who will be directly impacted are involved. You don't need a large and complicated scheme, and it is dead simple to implement, comparatively. You only have to trust the people involved, not swathes of departments.
Because, eventually, they'll leak. Security and intelligence agencies have lost keys and hacking tools and top security documents.
And when they leak, the surface for what you can exfiltrate before the keys are rolled over will be _everything_. If you can roll over the key.
Microsoft leaked their Secure Boot key once, and it wasn't possible for them to fix that on all of their devices. Whilst that isn't a particularly concerning leak, that changes if what it was protecting becomes someone's personal information.
To protect against a leak that compromises everyone, everywhere, you'd want the government keys to be unique per site and company. Which would also incidentally make securing those keys harder and lead to more leaks.
And that's before you question whether the people with access to the keys are actually trustworthy and won't do what law enforcement have done in the past, like stalk their ex.
It's a no-win situation.