It was found by static analysis FIVE years ago, but never re-found? That is really odd. I thought people were running every kind of commercial scanning tool on OSS for free most of the time.
Sure, you can get coverity scans for free for open source tools. Most open source maintainers I know aren't eager to hear ideas how they can have more work maintaining their software.
They depend on GPG but GPGME is something else. The latter is a wrapper around GPG mostly intended for email software.
Is is not even close to being as widely used as GPG itself, packaged and distributed as a separate piece of software, so it's fully understandable if big Linux distributions such as the ones you mention don't as much time on it they they would on, say, GPG itself.
Frankly, this could have been a problem with Coverity itself (particularly the UX). As someone who has used Coverity in professional setting, keeping tab of issues was hard. Or maybe it was our coverity deployment that sucked.
As a contributor to an OSS project that uses Coverity, yes, we do. The problem is that for a lot of these projects the first time you run any SAST on it you get about 5,000 results. A common approach for commercial users for dealing with this is that you fix all the critical ones and then mark the rest as the baseline and only fix new stuff beyond that point, because you don't have the budget or time to fix all 5,000 issues. For OSS I don't know the approach there, although in our case with security-focused devs with OCD we fixed everything before continuing.
Likewise, a maintainer of a project that uses Coverity. It is very useful, and I am thankful to them for generously making it available to us for free. But I agree it is a bit noisy with results. E.g., many of the reports will mark code as "dead" when it is used on other platforms.
Red hat was the standard suggestion when I started using Linux in 1998. I personally switched to Debian in '99 and it's still my go-to when I'm not using a *BSD.
When Ubuntu came on the scene it quickly became a dominant choice for normies and non-tinkerers, and I feel like this sort of lifted prominence for debian too.
In my personal experience, I think both of you are right in different bubbles (for lack of a better word). If you worked in commercial unix-like tech systems since the early 2000s (meaning your company/business was tech focused), you would consider RH to be the canonical distro. There is a reason all the newer distros marketed towards commercial use (Alma, Rocky, Cent, Core, Amazon, Azure) are all Fedora/RH derivatives.
If you have been mainly OSS or used unix-system in a non-tech business, then you would consider Debian/Ubuntu to be the more canonical distros.
Yeah I use Ubuntu and Fedora and the only thing I don't like about my Fedora desktop is that more desktop software is only available for Ubuntu/Debian rather than Fedora. Probably only simply because Ubuntu is more popular. (I have used alien in the past to some success in this regard to get software running on Fedora that was only built for Ubuntu.)
More on alien:
> alien is a program that converts between Red Hat rpm, Debian deb, Stampede slp, Slackware tgz, and Solaris pkg file formats. If you want to use a package from another linux distribution than the one you have installed on your system, you can use alien to convert it to your preferred package format and install it. It also supports LSB packages. [1]
This vulnerability was discovered by Johan Carlsson and we were informed by GitLab on 2026-11-12.
Initially we set 2026-11-25 for a coordinated disclosure;
due to the new Ruby2Shell bug GitLab had to update earlier
and thus we also disclosed the GPGME bug on 2026-11-20.
I guess pretty important for GitLab since it supports signed commits, code execution in that path is pretty bad.
AI is doing quite well at automating the escalation of small memory bugs (which can usually theoretically cause RCEs) into actual RCE exploits. This bug has a similar nature to php2shell.
I was too busy putting out the node2shell and others to look at the PHP vulnerability - was that one an easy “get shell” exploit or mostly theoretical one like the Node exploit where an attacker has to know the exact memory layout to else cause a crash…
It must mean something to "assume the worst". If you always assume the worst, you also never do it.
If an RCE exploit can't be ruled out, that's bad, but would be clearly worse in every way if an RCE exploit was public.
An LLM can aid in designing an actual exploit, but if it's not possible then it's simply not possible. Oftentimes we are lucky and exploits turn out to be impractical or require configurations never found in the wild.
If someone opens an encrypted email that came from an attacker using GnuPG and code is executed on his machine as a result, that is still remote code execution.
Are you suggesting the existence of exploits outside of the definition of "attacker controlled"?
If not, and all exploits are attacker controlled, it sounds like all exploits are RCE.
EDIT I guess I have to weaken the last condition since we haven't proved all exploits are "code execution", but we have at least proved all exploits are "remote".
The implication you're missing is remote execution of [ arbitrary and attacker provided ] code. An example of an exploit that doesn't qualify as RCE is a web server providing read only access to files it should not.
For example, if an attacker can remotely trigger only a specific preexisting program (perhaps sudo rm -rf /) and if the attacker is unable to directly escalate from that to full blown arbitrary code execution then that vulnerability would also not qualify as RCE.
In the gnupg case, if an email can contain arbitrary code that gets executed on the target system then you have RCE.
And if I leave a USB with a malicious executable in a parking lot, and someone runs it, it sounds like I'll have a remote code execution vulnerability too. Sounds like a very low bar, I guess RCE must be a basic, "everything gets this marker" thing.
Well, yes? That is a long established and very well known social engineering attack that indeed results in RCE pretty much by definition unless various technical countermeasures have been put in place by IT. This is precisely why USB (and most other) ports are physically removed in sensitive environments and why any competent operation will usually have a cryptographically based whitelisting approach when it comes to executing binaries and scripts.
The bar is simply "an attacker is able to cause the execution of arbitrary code without being physically present". I'm not clear what it is that you find so bothersome about this exceptionally straightforward criteria.
You're calling "trick someone into running an executable" an RCE? I can work with this definition, but like I said, it makes this a very low bar, very valueless definition. Everything is RCE. Which is why it's hard to answer my question "what _isn't_ RCE".
If your threat model includes an employee being tricked into running an arbitrary executable then yes that's RCE. If that's not part of your threat model then it doesn't constitute an RCE vector for you.
How common or uncommon a particular class of exploit is has absolutely no bearing on such criteria - these things are entirely functional in nature. The value of such definitions is in clearly delineating different concepts relevant to threat modeling. When an attacker has pwned you and is executing arbitrary code on your system are you going to object that "well ahckchually what you did doesn't qualify by my personal definition because ..."? Or perhaps you'll belittle them for using what you see as a pedestrian attack vector?
> Everything is RCE. Which is why it's hard to answer my question "what _isn't_ RCE".
You're intentionally being obtuse. I already provided examples. An SQL injection for example does not (at least typically) lead to RCE.
> You're intentionally being obtuse. I already provided examples. An SQL injection for example does not (at least typically) lead to RCE.
You've described something that's not "code execution", and I've agreed several times. I'm making fun of the fact that a "local privilege escalation" is turned into a "remote" attack if it's emailed and run/executed/parsed, much like an email signature.
The problem with that analogy is that it's not a vulnerability, that's how running executables from a USB stick works. No one expects importing a GPG public key to execute untrusted code.
I'd label it as Arbitrary Code Execution, since it requires the target to initiate the interaction. RCEs are ACE vulns where the target responds to an attacker-initiated interaction.
No it doesn't. The announcement email is reasonable in that it says a RCE "can't be ruled out", but whoever editorialized the submission title here to "RCE in Gnupg" was clearly going for clickbait.
It makes sense to assume the worst, then try to achieve it. And only if you achieve it so you tag it as such. You can't announce vulnerabilities to the world based on vibes of what it might do....
"Yeah this is a classic double free. We can't think of a way to transform it into a RCE or something like that, we've also failed at having LLMs do it so far. but in the current climate it could happen so we humbly ask the tag of most dangerous vulnerability please. Not just this disclaimer. We want 9.9 CVE that says RCE!!"
The mail says "Exploiting this bug to mount an Remote Code Execution attack (RCE) can't be ruled out.".
They didn't write this just for fun. gnupg usually downplays security issues.
The attacker would of had to have successfully managed to get the victim to import a malicious key with a super long userID before that. The signed message would just be the trigger.
The trigger, whatever it was, would have to not just cause the super long userID to be displayed, but would somehow have to affect the target's display code in a meaningful way to do some sort of exploit.
We are having this debate right now. So it's far from over. Without PoC it's unclear if it's RCE or just overhyped bullshit. See PHP or Nginx "RCE" which required specific crafted settings for it to even work. A video is not a "PoC".
If you believe an LLM can generate an exploit for this - feel free to post one.
It's more of a "made public" a year later, you have no idea if anyone exploited it in between.
At first I thought the whole Linux everything is a CVE was kind of annoying, but over time, especially with automated exploit finding it really has shown to be in front of what was about to happen.
>At first I thought the whole Linux everything is a CVE was kind of annoying, but over time, especially with automated exploit finding it really has shown to be in front of what was about to happen.
I support that, not the CVE part, because that one is annoying paperwork, but the "every bug is a security bug". What I don't support is overhyped panic spreading because something might have an RCE, without proof.
GPGME seems to stand for "GPG made easy" (whatever that might mean).
I have 2 library packages with such name installed on my Ubuntu laptop. They have been pulled in by podman and by libreoffice.
This doesn't seem to be the standard package signing mechanism used by many Linux distros. So yes, many people might have it (if the libreoffice path is affected). But standard package installation should not be.
(Just from my quick analysis. Not an expert in the area and did not dig deeper.)
What are practical attack vectors for Linux workstations and servers?
The masses of internet users don't know what GPG is. I'd caim the majority of software developers don't have a GPG key and never explicitly verify any signature.
But packages of many distros are verified using GPG. Is the vulnerable component involved or not?
> But packages of many distros are verified using GPG. Is the vulnerable component involved or not?
Debian uses it last time I was on a Debian system And I think all Debian based distros that don't modify apt will essentially use GPG to verify the packages as well.
I think it's one of the those programs that pop up in a lot of different things. Personally I use it a fair amount and even use the Symmetric encryption features for some files and for personal password management.
I do think developer are moving away from it thought I agree (and for various reasons that are good).
SPARK Ada does exist since the late 1980s (although not open source for a long time) and does solve a lot of security problems that plague C/C++.
But for use-after-free, from the beginning, the solution was not to have pointers in the language at all; Ada does have pointers and Ada.Unchecked_Deallocation.
Obviously dynamic memory allocating does have useful applications, so after Rust 1.0 was released, SPARK Ada added pointers and an explicitly Rust-inspired ownership model.
the pointer and heap manipulating points are very fair points indeed. i just hate when people act like rust is a totally new concept. (heck not to mention stuff like eiffel!)
shrug, depends on which industry exposure you have. but ill concede w the recent mandates and climate they're moving to rust from ada/spark and c/c++. one could argue if nothing more than the improved popularity and tooling (except compile times god damn)
I think it did. Borrow checking and ownership. The last influenced access types in SPARK. Besides that, Rust probably is better suited for application with threads and dynamic memory allocation.
It's not really helpful to suggest that a thing that has a fraction of the usage and interest of the thing with a discovered security vulnerability is the ultimate solution to the problem. You still need to fix the thing that is being used. The less popular thing might actually be worse than the popular thing.
The current title is "RCE in Gnupg". A more appropriate title would be "Double free produces posible RCE in GnuPG". The non-editorialized title would also be better because it does not claim there is an RCE that is not known to exist.
"Probable RCE in GnuPG clients" would be appropriate.
When someone posts a Claude POC, then the situation is different.
Not even that, just "double free" would be a fair title. Every double free is a possible RCE using this logic. For tens of years we had "double free" vulnerabilities. Calling each and every of them "RCE" makes that word meaningless.
> For tens of years we had "double free" vulnerabilities.
For tens of years we didn't have LLMs. Times have changed - a double free in a path exposed to attacker controlled input should be assumed to lead to RCE.
Sure, a double free in your calculator app probably does not provide a realistic avenue for RCE.
"when it was noticed by a Coverity scan:"
It was found by static analysis FIVE years ago, but never re-found? That is really odd. I thought people were running every kind of commercial scanning tool on OSS for free most of the time.
Sure, you can get coverity scans for free for open source tools. Most open source maintainers I know aren't eager to hear ideas how they can have more work maintaining their software.
Is is not even close to being as widely used as GPG itself, packaged and distributed as a separate piece of software, so it's fully understandable if big Linux distributions such as the ones you mention don't as much time on it they they would on, say, GPG itself.
When Ubuntu came on the scene it quickly became a dominant choice for normies and non-tinkerers, and I feel like this sort of lifted prominence for debian too.
Ubuntu is the Canonical distro, but that's unrelated.
If you have been mainly OSS or used unix-system in a non-tech business, then you would consider Debian/Ubuntu to be the more canonical distros.
IMO, RH is just better at security, integration, and much much more.
More on alien:
> alien is a program that converts between Red Hat rpm, Debian deb, Stampede slp, Slackware tgz, and Solaris pkg file formats. If you want to use a package from another linux distribution than the one you have installed on your system, you can use alien to convert it to your preferred package format and install it. It also supports LSB packages. [1]
[1]: (https://manpages.ubuntu.com/manpages/bionic/man1/alien.1p.ht...)
You can find the GitLab Omnibus related patch here: https://gitlab.com/gitlab-org/omnibus-gitlab/-/commit/279ac3...
> Exploiting this bug to mount an Remote Code Execution attack (RCE) can't be ruled out.
If an RCE exploit can't be ruled out, that's bad, but would be clearly worse in every way if an RCE exploit was public.
An LLM can aid in designing an actual exploit, but if it's not possible then it's simply not possible. Oftentimes we are lucky and exploits turn out to be impractical or require configurations never found in the wild.
In all seriousness, “remote” in this case means attacker controlled.
If not, and all exploits are attacker controlled, it sounds like all exploits are RCE.
EDIT I guess I have to weaken the last condition since we haven't proved all exploits are "code execution", but we have at least proved all exploits are "remote".
For example, if an attacker can remotely trigger only a specific preexisting program (perhaps sudo rm -rf /) and if the attacker is unable to directly escalate from that to full blown arbitrary code execution then that vulnerability would also not qualify as RCE.
In the gnupg case, if an email can contain arbitrary code that gets executed on the target system then you have RCE.
The bar is simply "an attacker is able to cause the execution of arbitrary code without being physically present". I'm not clear what it is that you find so bothersome about this exceptionally straightforward criteria.
How common or uncommon a particular class of exploit is has absolutely no bearing on such criteria - these things are entirely functional in nature. The value of such definitions is in clearly delineating different concepts relevant to threat modeling. When an attacker has pwned you and is executing arbitrary code on your system are you going to object that "well ahckchually what you did doesn't qualify by my personal definition because ..."? Or perhaps you'll belittle them for using what you see as a pedestrian attack vector?
> Everything is RCE. Which is why it's hard to answer my question "what _isn't_ RCE".
You're intentionally being obtuse. I already provided examples. An SQL injection for example does not (at least typically) lead to RCE.
You've described something that's not "code execution", and I've agreed several times. I'm making fun of the fact that a "local privilege escalation" is turned into a "remote" attack if it's emailed and run/executed/parsed, much like an email signature.
If RCE excludes Stuxnet, then maybe we should come up with a phrase that's less wrong.
"Yeah this is a classic double free. We can't think of a way to transform it into a RCE or something like that, we've also failed at having LLMs do it so far. but in the current climate it could happen so we humbly ask the tag of most dangerous vulnerability please. Not just this disclaimer. We want 9.9 CVE that says RCE!!"
This is ridiculous.
It's still a critical bug, just would not call it a prime example of rce.
I thought we reached the point a few years ago where the assumed probability this can be done is 100% ?
The trigger, whatever it was, would have to not just cause the super long userID to be displayed, but would somehow have to affect the target's display code in a meaningful way to do some sort of exploit.
If you believe an LLM can generate an exploit for this - feel free to post one.
Quit writing memory free errors in software. Prove your software is safe, not prove that it's not exploitable.
At first I thought the whole Linux everything is a CVE was kind of annoying, but over time, especially with automated exploit finding it really has shown to be in front of what was about to happen.
I support that, not the CVE part, because that one is annoying paperwork, but the "every bug is a security bug". What I don't support is overhyped panic spreading because something might have an RCE, without proof.
Then someone should do the needful and ask one to generate an exploit.
Otherwise it's just a bunch of hot air to stir up paranoia.
"Put your money where your mouth is", as the saying goes.
I have 2 library packages with such name installed on my Ubuntu laptop. They have been pulled in by podman and by libreoffice.
This doesn't seem to be the standard package signing mechanism used by many Linux distros. So yes, many people might have it (if the libreoffice path is affected). But standard package installation should not be.
(Just from my quick analysis. Not an expert in the area and did not dig deeper.)
The masses of internet users don't know what GPG is. I'd caim the majority of software developers don't have a GPG key and never explicitly verify any signature.
But packages of many distros are verified using GPG. Is the vulnerable component involved or not?
Debian uses it last time I was on a Debian system And I think all Debian based distros that don't modify apt will essentially use GPG to verify the packages as well.
I think it's one of the those programs that pop up in a lot of different things. Personally I use it a fair amount and even use the Symmetric encryption features for some files and for personal password management.
I do think developer are moving away from it thought I agree (and for various reasons that are good).
- Build Date: 2026-11-21 13:29 UTC
- Signature Date: 2026-11-21 13:31 UTC
- Last Updated: 2026-11-22 09:13 UTC
It's specifically in 'GPGME' though, so more limited scope I think.
Without digging into exactly how pacman uses it, I'm wondering if the cautious move is not to update anything before first updating it...
A lot of people use it especially certain high value targets on darknet drug markets who might miss this.
A lot of projects do still use it for signing projects though many have opted for simpler solutions (or unfortunately no signing).
I think Debian still uses it (not sure about ubuntu) in apt.
Sure, GPG gets used for signing packages and that COULD potentially lead to a supply chain attack IF this is not quickly patched.
But for use-after-free, from the beginning, the solution was not to have pointers in the language at all; Ada does have pointers and Ada.Unchecked_Deallocation.
Obviously dynamic memory allocating does have useful applications, so after Rust 1.0 was released, SPARK Ada added pointers and an explicitly Rust-inspired ownership model.
"Verification of Programs with Pointers in SPARK", 2020 https://dl.acm.org/doi/10.1007/978-3-030-63406-3_4
"Proving heap-manipulating programs with SPARK", 2021 https://av.tib.eu/media/53376
https://alb-lang.org/wiki/SPARK_Ada
Clearly it's bringing something to the table, even if not technical.
The correct title is: GnuPG Security Advisory for GPGME
"Probable RCE in GnuPG clients" would be appropriate.
When someone posts a Claude POC, then the situation is different.
Not even that, just "double free" would be a fair title. Every double free is a possible RCE using this logic. For tens of years we had "double free" vulnerabilities. Calling each and every of them "RCE" makes that word meaningless.
For tens of years we didn't have LLMs. Times have changed - a double free in a path exposed to attacker controlled input should be assumed to lead to RCE.
Sure, a double free in your calculator app probably does not provide a realistic avenue for RCE.