There is nothing that stops tech firms doing exactly what they're being asked to do. Every claim otherwise is obfuscation because tech firms don't want to do it, mostly because of their internal internationalist politics where they don't want to be forced to pick sides and tell some governments "sorry, we're Anglos who choose to give the US/UK governments special access that you don't get because we're better than you". It's an example of the first kind of no, not the third kind.
End-to-end encryption is a vipers nest of false claims like this. There's lots of ways to implement such policies, like this: for each message that's being encrypted, you encrypt it under a per-message key which is then itself encrypted under the recipient's key, and also a police (public) key. The servers forward the messages to the police so they can decrypt them. If the decryption fails too often (hacked client) then that user is denied access to the network. Yes yes I know that WhatsApp/Signal and friends use a more complex protocol, that description is a simplified textbook example, but the argument doesn't change.
Cryptography is a very flexible set of tools. They can easily be used to achieve complex security goals, like empowering some parties whilst disempowering others. The resistance to doing this is legitimate and I even agree with it, but it's also political and not technological. When politicians push back and insist that their police should have access to WhatsApp, and get told it's impossible, well they are not all stupid and correctly conclude they're being bullshitted. Indeed some of the MPs in the UK have computer science degrees.
Fact is, buzzword salads can be used to baffle people and get them to agree with you even if you're wrong. Technologists are especially tempted to abuse them when they want to say "no" to make a Type 1 No seem like a Type 3 No. Researchers do the same thing all the time, your complaints about grant funding are as old as the hills. Honest specialists speak clearly even when they might benefit from speaking unclearly.
Btw, end-to-end encryption by its very definition, means that only the sender and receiver kan decrypt it. Your scheme is basically saying that the police should also be a receiver of all messages...
Remember that the encryption is being done client side by apps these networks control. E2E is therefore sort of fake to begin with because WhatsApp is not only the servers but also the client. You can't mix and match, so you have to encrypt messages using software provided to you by the "adversary". E2E encryption is therefore more of a tool to control bad insiders and negotiate with governments than encryption as conventionally understood.
Also remember that tech firms run the public key directory. Almost nobody verifies public keys, and even if they did, they're doing so with apps controlled by the tech firms so you can't know the verification is done properly anyway. And the keys can change at any moment, with your own way to know it's happened being UI controlled by the tech firms.
Still, even if clients and servers were separate, nothing stops clients from encrypting messages using a well known government public key and attaching that along with the e2e encrypted version.
The messaging company can embed the police's encryption key in the app but not have possession of the corresponding decryption key.
> exceptions to privacy have always led to a slippery slope
That's a reasonable argument. But to the GP's point, thats not a technical argument. Its just another argument that the policy is bad for normal "bad policy" reasons. It has nothing to do with the math.
Once three people have access to a secret, it isn’t secret anymore. Once hundreds of thousands of police officers have access to the private key, it will leak, and everyone will be able to read these messages.
Again, do you see how this is not a technical argument? It might be a good argument. But its not an argument about the math or the computer science. "We can't trust the police" is a social argument, not a technical argument. A math or CS degree will not help you understand this argument.
Anyway, why would the decryption key be in the hands of "hundreds of thousands of police officers"? Especially when the decryption key itself is useless without access to the encrypted messages themselves. If this were implemented, its much more likely that the police would build themselves a web portal or something through which they could access people's WhatsApp messaging logs. The crypto could all be handled on the data portal backend.
A much stronger argument against this sort of thing is the governmental slippery slope argument. If the UK police gains capabilities like this, you bet every other country will (reasonably) demand similar access. Apple / Meta would have to decide which police / security departments to work with, and thats a very complex problem. Who do you trust? Hungary? Bulgaria? Russia? Iran? Egypt? China? Brazil? Where, exactly, is the line? And should access be revoked after a coup, like in Niger?
Its much easier to just refuse all governmental cooperation. It protects your brand. And makes it much simpler to justify refusing access to police departments who you don't trust.
Arguing that the myriad local police departments of the United States in particular do not have the security posture required to keep access to a data portal secure is a technical argument against government-backdoor encryption.
> “Work it out” is generally a demand to invent new mathematics, but sadly, mathematics doesn’t work like that.
E2E + a backdoor for a particular police department isn’t the sort of bad idea that requires new mathematics.
> For the last 25 years, engineers have said ‘we can make it secure, or we can let law enforcement have access, but that means the Chinese can get in too” and politicians reply “no, make secure but not for people we like”.
Insecure police departments will inevitably leak the backdoor keys. It's not possible to limit who can use decryption keys based on who they are and not just possession of the keys under our current understanding of how encryption works. If you assume that the police will never leak keys then sure it's easy. But arguing about whether or not social factors like police department computer security is good enough to safely store keys is a technical argument about this technical problem.
It's quite depressing how many people here don't seem able to deconflate "I don't believe police can be trusted with this power" from "it is mathematically impossible to give it to them".
This is a nice clear example of how experts talk themselves into lying to the public for the greater good, as we saw so often in the past :(
And serverless literally means "without servers", and yet...
Point being, scope is a free variable. "E2EE" that's managed by a central server is already stretching it, yet people accept it. They'll mostly accept excluding law enforcement out of the scope of eavesdropers "E2EE" protects you from too.
The protocol you have described is not end-to-end encryption by that definition. Sure, it's possible to do this kind of escrow encryption, but that's not what Meta and Signal are selling, and it's definitely not what I'm buying.
To be clear, I think this is a horrible idea, but it’s not technically impossible.
It's perfectly viable to create a messaging application with encryption that isn't e2e. Sure, it's a ridiculous thing that no democracy should demand, and will currently destroy any brand that attempts it, for good reason. It also won't be effective against any mildly persistent threat (not much of a change over our current implementation, really). But it's something perfectly viable to create.
It's worth repeating this because tech firms have made the definition so confused, but encryption was developed to let you use a trusted device to communicate over an untrusted medium (radio). If your trust in the communications medium is the same as your level of trust in the device, which for so-called "E2E" messengers it is, then the whole system doesn't make any sense.
What Meta/Signal sell is kind of a smokescreen because they control both the clients and the medium and the key directory too, so nothing is really limited. They can update the logic at any moment to disable the encryption for you, the person you're talking to, or everyone, and nobody would ever know. They can also update the client to upload your private key if you're being specifically targeted, or use a weak RNG or suppress a key rotation notification or any one of a million other things. In fact, they might have already done that without anyone noticing. I pointed out in other posts that they already undermined one of the most basic properties of a modern cryptographic system (that the adversary can't tell if you're sending the same message twice) and they did so for typical government-type reasons of controlling rumors and misinformation, as they see it.
For E2E messengers to work conceptually they'd need to allow arbitrary third party clients, so you could choose to trust that client and then use the WhatsApp/Signal networks even though you don't trust them. Or at the very least, they'd need a very sophisticated and transparent auditing programme. They won't do either of those things.
By not sending a copy to the police, you can guarantee that those things won't happen.
... or because a police insider is threatened and/or hit repeatedly with a $5 wrench. One of the most effective decryption tools.
Police can store keys inside HSMs. Tech firms can give training and develop software stacks to ensure secure workflows if they want to.
You can also use a targeted approach. Allow the police to request intercepts of specific phone numbers. Tech firm rotates the public key being served to the clients, replacing it with a police key. They add a special flag that says "don't put any sign this happened in the UI". Now police can intercept just that specific phone number.
There's lots of ways to do this at different points in the security/usability spectrum, it's all just an engineering problem.
Also, bear in mind that police and governments won't find this argument convincing because they can already access every other form of electronic communication and in reality mass leaks of private data have never been an issue.
This is just wildly untrue. The 2015 OPM hack (in which an APT stole security clearance records) was a mass spill of private data. The Shadow Brokers leak put the NSA's most sensitive tools on the public internet for anyone to download. And of course we might never hear about collected intelligence that had been purloined by an adversary in the course of an offensive cyber operation.
And I know about the Shadow Brokers. Not personal information.
"Tools and methods" are in fact considered more sensitive than any personal information in the IC. The system you're proposing presumes the existence of at least one unhackable organization, and I regret to inform you that there is no such thing.
Firstly, nothing I've proposed demands unhackable organizations. That appears to be a requirement you invented. No security system presumes that! The so-called E2E encryption systems are hackable via several different organizations today: you could hack Meta client teams and insert code into the next releases, you could hack Google/Apple and tamper with the code as it gets shipped via the store, you could hack a phone OEM and insert a backdoor into the devices themselves.
Secondly, if you think government agencies would get hacked more often than tech firms then you may be right, but that's also irrelevant to any points being made here. The goal here isn't to design a perfect system, it's just to point out that the claim that no system can exist at all isn't true. Responding to that with "but your comment doesn't contain a full design doc for a system I personally judge as perfect" isn't going to get us far. Governments don't care if the system is perfect, right? They're OK with some leaks from hacked police departments.
Thirdly, I haven't even been making concrete proposals! Just pointing out how cryptography works as examples. If I was hired to implement these requirements tomorrow I wouldn't do things directly in those ways, they're oversimplified, hence the references to textbooks.
Sure, iCloud backups and photos are technically readable (they're not under e2ee) but policies and procedures exist to allow law enforcement to access them as well.
What you're asking for is that e2ee be eliminated in favor of a process-heavy solution. But at the end of the day, any human-dependent process can be broken by social engineering and a lack of vigilance.
Yes, an e2ee session can be broken by shipping a hacked client update to an endpoint. But that's much harder to accomplish, human- and process-wise, than obtaining the right key from the right person.
Key leaks are also relatively easy to prevent because secret keys are not relevant to investigations, so they never have to leave the interception framework and handed to a human operator. Sure, the infrastructure could be compromised, but that's already an extremely severe issue if you just have metadata in it.
France, 1940, religious census data.
It's not at all, which is why your solution ideas miss the mark completely.
It is fundamentally a "humans can be corrupted" problem.
Everything you describe is a backdoor of various forms. Backdoors have nothing that prevent abuse other than triple pinky promise to only spy the bad guys (by whoever gets to define who is the bad guy, which changes over time).
Whenever you have a system where people have access to bypass it, you'll soon enough have some of those people be corrupted to abuse it, or ordered by higher ups to abuse it or threatened by various way to abuse. Since people are involved, there is a 100% chance this will happen.
As long as corruptible humans are in the loop (i.e. as long as humanity exists) the one and only way to avoid corrupted abuse of backdoors is to not have any backdoors.
Tech firms also know that programmers can make mistakes whilst implementing complex cryptography, or even be corrupted, yet this is not itself an argument against implementing cryptography!
To repeat once again, we're not debating the ethics of E2E encryption here. Please don't waste time trying to convince me that E2E is encryption is a good idea, because (if it was real and worked) I'd agree with you! But your argument is a Type 1 No by the scheme presented in the article. It is a "we'd really rather not" social argument.
The problem our industry has is typified by the article. Too many tech people argue that giving police access to WhatsApp encryption is actually a Type 3 "it can't be done for fundamental physical reasons so it doesn't matter who demands it" problem, but that isn't true. Remember that governments don't care about E2E encryption whatsoever in any way. They would much rather ban it as a source of unnecessary problems. If tech firms claim they can't turn it off completely, they're obviously lying and that will just enrage governments. If tech firms claim they can't keep it whilst providing targeted access, governments don't care about that either. After all, email isn't end-to-end encrypted, nor is SMS, nor are normal phone calls, nor are letters. Why should WhatsApp be any different?
In reality it actually is possible to design a system that stops people with only server-side access to WhatsApp reading messages, whilst still breaking if the clients are compromised, and which allows police to have targeted levels of access without any risk of universal master key leaks. There are lots of ways to do that. You can use secure enclaves, zero knowledge proofs, or more exotic algorithms. But it's also not really relevant to the point I'm making, which is about the No Type being presented to governments. There was surely a better example that could have been chosen for a Type 3 No.
This is changing the goalposts.
Giving police access to whatsapp chat is trivially easy. But that's not the question.
The pro-surveillance people say "Give the good-guy police (whoever they are) access to everything and keep it secure from the Bad Guys Whom We Oppose (whoever they might be this week)". That one is indeed impossible due to the laws of information theory.
> without any risk of universal master key leaks
You're looking at a technical problem when this is not that, it is a humanity problem.
It's relatively easy to avoid e.g. master key leaks. That's an irrelevant implementation detail. What matters is that if some set of people have unfettered access to bypass all protections, then all the Bad Guys will also have that access soon enough because you can't keep people from getting corrupted/threatened. No matter how hard you wish, you can't. Humans are like that.
So, is this impossible due to human nature or the "laws of information theory"? Which is it? And if the latter what "law" are you thinking of, exactly? Can you name these laws?
Here's the problem: it's neither impossible mathematically nor practically. Remember the kerfuffle over the NSA's backdoored ECDRBG algorithm? That was a very pure textbook example of what's possible, it would have allowed the NSA and only the NSA to decrypt TLS streams using it. According to you that would have been physically impossible due to violating some sort of law, but it's not. Cryptographers know how to construct such systems, there are many ways.
But the existence of such solutions doesn't even matter. Lawful intercept abilities have existed for ages, governments will happily accept a solution of just turning off E2E encryption entirely, and the possibility that governments or tech firms will get hacked doesn't bother them because that's a transient problem that can be made hard by throwing money at it.
They also don't care about corruption because the countries demanding this have low levels of it. Governments are the original experts in corruption, you might say, and have evolved a lot of different mechanisms to fight it. Finally, remember that all these systems are already hackable or corruptible. Pay off someone who works on the WhatsApp mobile app team and unless Meta's internal controls detect it, it's game over.
Yes. It's simple, if you send information in a recoverable way to another entity, they can recover it. If that entity involves humans, they can and will with 100% certainty be corrupted or threatened to obtain the information improperly.
> Remember the kerfuffle over the NSA's backdoored ECDRBG algorithm?
Amusing that your example is a counterexample to your thesis. Exactly. Backdoors never serve only their would-be masters. That's the impossible part.
If this is not blindingly obvious by now, I fail at being able to explain it better.
So it's not a counter-example and your amusement is misplaced.
My old boss and I were recently laid off because my company (Livingston International) has been doing massive layoffs, because the company has been losing clients and money over repeated phishing attacks. We always shat on the assumed tech-illiterate person or persons responsible for us now having to sign in multiple times per DAY with MFA. Right before we left he mentioned this suspicious email he got the other day… and he clicked on it. It was one of those internal “we got you! Be careful next time!” emails. Come on man.
https://www.cisco.com/c/en/us/td/docs/switches/lan/catalyst6...
It says it's from 2015, but the Catalyst 6500 product line is much, much older. And quite a few companies sold both mediation devices and encryption software.
All these problems exist with access to metadata, so police and service providers already need to address them today. I'm not saying that this isn't challenging (maybe it is, I don't work in this field), it's just that handling content (especially text message) within the existing framework wouldn't be such a massive change because of the existing infrastructure and procedures.
In countries with higher corruption levels than the US data/metadata from “lawful” access is sold on a black market. I don’t think the problem of keeping data safe solved in a future proof way even in the first world countries and if encryption will be compromised sooner or later data or even keys will be sold/stolen/abused.
The hard part here isn't key leaks. System critical keys are commonplace in our society and virtually never leak, exactly because they don't tend to leave dedicated hardware. For example e-Passports are signed with long term government keys, they don't leak all the time. The US DoD runs a large scale private PKI, no problem. Our society has even got pretty good at physically giving people private keys in such a way that they still can't access them: credit cards, SIM cards, games consoles.
Even if you assume HSMs are constantly getting ransacked, governments don't care. They don't even necessarily want to have their own key management to deal with at all. A web portal that employees log in to, type in a phone number and then see the logs is perfect for them. Make it be dedicated hardware supplied by Facebook itself if you want, with login systems as secure as they use for their own employees. Governments just do not care about these details. Type 1 Nos, to use your lingo.
The hard part of such a system is defining your precise security goals and then implementing it in ways that all the goals are met simultaneously. So called "E2E encryption" isn't really, we all know that, so there's lots of flex to define systems that meet the same goals in different ways especially if you're willing to roll with good-enough type solutions e.g. assume a trusted client (which e2e messengers do already for things like their forwarding counters).
However this goes both ways. People demanding a solution think that you could make something investigable while keeping it completely airtight to adversaries or abuse. There is no escaping the fact that this is mathematically impossible in terms of how risk works. You can compromise security in favor of investigability, or you can improve security at the cost of investigability. And it's also important for the lawmakers to understand that each compromise is not gradual. It's drastic. If you went from 1 party having a key, to 2 parties, you've probably doubled vulnerability surface. If it's 3 parties, and one of those parties is an organization with lots of employees, you've probably exponentially increased vulnerability surface by orders of magnitude. This math does apply here.
One experience was a trip to a crypto conference. Many booths were making unsubstantiated, impossible claims. If you tried to ask how, they couldn't answer. Like, someone would say they made transactions instantaneous, and when I asked how they solved the problems of unpredictable networks, they'd have no answer. It got pretty frustrating. I had to dig through a few claims like these only to be convinced that they're selling snake oil.
Another (less recent) experience was looking through a spreadsheet of government approved innovation/research grants. I couldn't believe what I was reading. Many of the entries seemed to make absolutely no sense, purposefully using buzzwords to sound smart, but having no meaning when unpacked. Buzzword salads. These are funded projects. And to get more money on round 2, all the "innovators" had to do is show any activity, which is very easy to fake.
So yes, tech charlatans. I'm a bit on the old school side of tech, and this gives me cognitive dissonance. I'm used to thinking of software devs/engineers as honest/creative/driven, but I guess this was always inevitable. Our field is very exploitable, because many people put their livelihood in tech, while knowing very little about it.