The PGP Problem: A Critique
You can read the revision of this article before the general rewrite but you should read this one instead. It is better.
The anti-PGP rant in question can be found here:
This is a pretty good rant. I feel like I am attacking a work of art with an axe here, but the rant promotes misconceptions that need to be addressed.
I only use Signal Messenger as a counter example in the following because the author of the rant uses it as an example of something that they think could be used as a replacement for encrypted email. I currently consider Signal Messenger one of the best in its class (2021).
The PGP Problem
Fortunately, if you’re not morbidly curious, there’s a simple meta-problem with it: it was designed in the 1990s, before serious modern cryptography.
The entire rant is basically about how OpenPGP is old and therefore bad and how new things, sometimes only vaguely defined, are good. So let's address this first.
If someone, while trying to sell you some high security mechanical system, told you that the system had remained unbreached for the last 20 years you would take that as a compelling argument. You would be unlikely to demand a newer design. Normally old designs that have stood the test of time are valued. Cryptography is based on mathematical/logical principles. Such principles don't age out on any sort of a schedule and are valued in some cases for thousands of years.
So the rant is contending something that goes against conventional expectations. Normally that would require some evidence and/or a good argument. The rant provides neither.
“PGP” can mean a bunch of things, from the OpenPGP standard to its reference implementation in GnuPG. We use the term “PGP” to cover all of these things.
Leaving out the context can make a rant flow better but quietly switching back and forth between fundamentally different issues of specification and implementation does not at all improve the chances of proper understanding.
For reasons none of us here in the future understand, PGP has a packet-based structure.
A long term standard must be extensible. Thus the fields can not be fixed and it must be possible to add new types. Here is the structure of an OpenPGP packet:
Explicitly defining the length at the start of the packet makes the result resistant to things like length extension/truncation attacks. The simplicity of this data structure makes the result resistant to attacks on message syntax.
There are at least 8 different ways of encoding the length of a packet, depending on whether you’re using “new” or “old” format packets.
This is true in a narrow technical sense but is misleading. Both old and new formats are variable length. The old format has a 2 bit field to indicate the number of bytes used for length. The new format uses a simple length extension based on the bit patterns of the first byte. This is a common technique. UTF-8 does the same sort of thing but no one is claiming that there are 6 (4) different UTF-8 representations.
The “new format” packets have variable-length lengths, like BER (try to write a PGP implementation and you may wish for the sweet release of ASN.1).
For part of my working life I had to implement low level protocols from specifications of various kinds. If I had of encountered the OpenPGP packet structure I would have considered implementing to be a relatively good time. The reader is invited to experience the overwhelming complexity of the OpenPGP packet structure. It is defined in section 4 of RFC-4880. The definition is all of 4 pages long. It includes code examples for each case of the length extension and a complete list of possible tags.
The most recent keyserver attack happened because GnuPG accidentally went quadratic in parsing keys, which also follow this deranged format.
This had nothing to do with the parsing of the packet format and it was not accidental. GnuPG undoubtedly did things the way they did for simplicity. That's a common theme in the security focused OpenPGP ecosystem. To make the attack practical the attacker had to add tens of thousands of signatures to the target key. Some had over a hundred thousand signatures added.
The actual system doesn’t get simpler.
Compared to what working public key based public key system? OpenPGP as a protocol is relatively simple.
Mired In Backwards Compatibility
If you’re lucky, your local GnuPG defaults to 2048-bit RSA, …
Which is wild overkill, but I don't think that the writer meant to imply that. RSA with 2048 bit keys is a perfectly reasonable and conservative default, particularly for an offline capable, stateless protocol like OpenPGP.
… the 64-bit-block CAST5 cipher in CFB, …
Here are the symmetrical ciphers GnuPG version 2.2.12 specifies by default, ordered from highest preference to lowest preference:
Cipher: AES256, AES192, AES, 3DES
CAST5 isn't even in the list, much less a default of some sort. CAST5 remains unbroken and would be quite suitable for use in a length limited, offline, stateless medium like email. CAST5 actually serves as an example of how old encryption methods can still be completely usable. CFB (Cipher Feed Back) is actually sort of awesome.
…that mixing compression and encryption is dangerous…
This comes from an attack on an online, stateful encryption protocol that involved information leakage due to compression. This sort of attack is not applicable to the sort of offline, stateless systems that OpenPGP is used for. See: Oracle Attack Immunity.
Mixing apples and oranges in this way is common. The rant fails to make this distinction in more than one place.
Here we are clearly talking about implementations…
There was a PGP usability study conducted a few years ago where a group of technical people were placed in a room with a computer and asked to set up PGP. Two hours later, they were never seen or heard from again.
There have been several encrypted messaging usability studies conducted over the years. They tend to use OpenPGP implementations rather than the encrypted messenger of the week for obvious reasons. They are usability studies. They are deliberately designed to be as difficult as possible (you hide all the documentation) in order to get results. If you wanted to compare the usability of a particular PGP supporting client to other clients you would have to make a different study.
In a usability study involving Signal1), 21 out of 28 computer science students failed to establish and maintain a secure end to end encrypted connection. The usability of end to end encrypted messaging is a serious issue. We should not kid ourselves into thinking it is a solved issue.
PGP begs users to keep a practically-forever root key tied to their identity.
Most people prefer to keep their identity indefinitely. That is why, say, the Signal protocol also has a “practically-forever root key tied to their identity”. It is inherent to the problem. I think the author might of been making an obtuse reference to forward secrecy and got a bit lost along the way.
The PGP cheering section will immediately reply “that’s why you keep keys on a Yubikey”.
… which would be a reasonable point. An air gapped system such as a Yubikey does significantly reduce the chance of a key compromise, both the “practically-forever root key tied to their identity” and the encryption key actually related to the forward secrecy issue. A user would prefer not to have any keys compromised. An offline encryption scheme enables the practical use of a portable air gapped system. Such a thing would not be practical with an online instant messaging scheme such as Signal Messenger. See: Encrypted Email is More Secure than Encrypted Instant Messaging
This section is based on a misconception. A stateless, offline capable system such as OpenPGP does not require the sort of authenticated encryption meant here and the Modification Detection Code (MDC) is not some sort of poor substitute for such an authenticated encryption scheme. For the details see the Authenticated Encryption article.
The PGP MDC can be stripped off messages –– it was encoded in such a way that you can simply chop off the last 22 bytes of the ciphertext to do that.
Well, sure, you could do that. An implementation would probably just stop with some sort of end of file/message/packet error. You obviously wouldn't end up with a valid MDC check so this is an odd thing to suggest.
To retain backwards compatibility with insecure older messages, PGP introduced a new packet type to signal that the MDC needs to be validated; if you use the wrong type, the MDC doesn’t get checked.
An application that required the MDC would obviously not accept an entirely absent MDC.
Even if you do, the new SEIP packet format is close enough to the insecure SE format that you can potentially trick readers into downgrading; …
We have a problem here. The juxtaposition of the non sequitur about chopping off the last 22 bytes makes it seem that that is all that is required to downgrade the MDC. Some digging reveals that this is actually quite difficult and has a very low chance of success(1 out of 32768)2). We can be charitable here and assume that the author simply did not do any research but this is very misleading.
Trevor Perrin worked the SEIP out to 16 whole bits of security.
This was wrong, but it was not Trevor Perrin's error. It turned out that the specification was wrong. Trevor Perrin was insightful enough to notice that the system described in the specification was vulnerable to this particular attack. The specification was corrected to what the implementations were actually doing and the vulnerability went away. This discussion was from the IETF OpenPGP standard mailing list3). If any actual MDC weaknesses had come from the discussion then they would have been resolved at that time. There is no reason to think that there is anything wrong with the MDC. This discussion was part of the process intended to ensure that the MDC is secure.
And, finally, even if everything goes right, the reference PGP implementation will (wait for it) release unauthenticated plaintext to callers, even if the MDC doesn’t match.
There is no such “reference PGP implementation”. OpenPGP takes the form of an IETF published standard. The Signal Protocol has no such associated standard. It appears you are supposed to read the high level overview and then transliterate the Java implementation.
Other people organize “key signing parties”.
Alas, this is something that doesn't really happen anymore. A key signing party is a good way to give people an intuitive grasp of cryptographic identities; something that would be valuable to the user of, say, Signal Messenger.
Ordinary people will trust anything that looks like a PGP key no matter where it came from …
In the same way that “ordinary people” will never bother to check their key fingerprints when using other systems. They will also ignore warnings that those fingerprints have changed. The problem is logically the same and OpenPGP is no worse off here (see the proceeding comment about the Signal Messenger usability study). PGP in a sense is better here in that it forces the user to comprehend the existence of a key in a way where it is intuitively obvious that it is important to know where that key came from.
Messages are (in normal usage) linked directly to key identifiers, …
It is important to note that this linkage is only to the recipient of the message. Normally in a messaging system the recipient has to be provided to allow the message to be routed to the destination. If someone were to come up with a system that made the recipient anonymous it would be little extra work to omit this linkage. Such usage is explicitly supported under the OpenPGP standard.
Further, a rather large fraction of PGP users make use of keyservers, …
They do? A PGP keyserver links an email address to a PGP identity. That identity might not be linked to a physical identity at all. If you did not want this linkage then you would have no reason to use a keyserver.
No Forward Secrecy
Reduced to the essence; forward secrecy is where you delete the encryption key protecting some encrypted data to prevent that key from falling into the possession of an attacker that already has that encrypted data. There is nothing preventing any system from doing that, even something based on the OpenPGP standard. For a practical demonstration see: A Demonstration of Message Burning Through Encryption using GnuPG.
Forward secrecy means that if you lose your key to an attacker today, they still can’t go back and read yesterday’s messages; …
Unless those messages are archived, which is normally the case with email and tends to be the default on other systems (e.g. Signal Messenger). If they get your private key then they pretty much for sure get your archived messages. Very few people are willing to go without a message archive so forward secrecy is unlikely to help in most practical cases of messaging.
In modern cryptography engineering, we assume our adversary is recording everything, into infinite storage. PGP’s claimed adversaries include world governments, many of whom are certainly doing exactly that.
Almost all email in transit (>90%) is protected by SMTP STARTTLS these days. So in practice those world governments don't have access to your PGP messages recorded off the wire anymore.
It’s theoretically possible to achieve a facsimile of forward secrecy using the tools PGP provides. Of course, pretty much nobody does this.
The interesting question here is: why not? It's easy to do in a technical sense and would not cause your correspondents to have to reverify your identity.
In 2020 a company called Cellebrite announced that they had a specific Signal Messenger data interpretation facility that would be useful in the event that their forensic box broke into the smartphone that Signal was running on4). This sort of attack can not reveal any messages archived off the network due to Signal's forward secrecy. Let's compare the end result to an encrypted email client running on the same phone using traditional passphrase protection:
|Archived Network Messages
|Messages Saved on Phone
So the encrypted email actually ends up providing a better result for the user. That is because it is possible to lock up the secret key material more securely in practice with an offline medium than it is with an online, always available, medium like instant messaging. It seems possible that people don't bother with forward secrecy for email because they perceive it to be secure enough already. Forward secrecy might not be worth the extra effort for that particular medium.
Please see the forward secrecy article for a somewhat more extensive discussion.
This section compares apples to oranges to kumquats. Categorizing the different things here would be so tedious that I will just give up and admit defeat. Let's move on.
I am reasonably certain that this section is based on a misapprehension. An offline stateless protocol like OpenPGP is not susceptible to downgrade attacks in the ways that online stateful protocols are. Negotiation is simply not possible for a medium that involves generating an encrypted/signed message/file and just sending it somewhere. See: Downgrade Attack Immunity
I do not agree that compatibility is a bad thing. OpenPGP is actually a good example of how to deal with backwards compatibility in a good way.
I have not been involved with the GnuPG source code enough to venture some sort of opinion on its quality. Presumably the author of the rant has. The GnuPG project treats side channel attacks as faults and fixes them as a matter of policy. As a result researchers will take the time to find them and submit them.
The 2018 Efail vulnerability was a result of it releasing unauthenticated plaintext to callers.
Since the EFAIL topic is now on the table, I can't help but bring up some related points. The much beleaguered OpenPGP modification detection code (MDC) reliably detected the EFAIL attack. The researchers were not able to overcome it. The also disparaged OpenPGP cipher feed back (CFB) caused those same researchers considerable trouble.
GnuPG is also effectively the reference implementation for PGP, and also the basis for most other tools that integrate PGP cryptography. It isn’t going anywhere. To rely on PGP is to rely on GPG.
The popular Thunderbird email client switched from GnuPG to RNP. Is RNP “effectively the reference implementation for PGP” now? I count 20 different OpenPGP implementations on the OpenPGP.org developer page5). Heck there are at least two active Rust implementations available right now6).
Comment not required here.