GPGME CVE-2026-129975 Double Free, RCE Not Ruled Out: Update to 2.3.1 Now

Rescana advisory card: GPGME CVE-2026-129975 double free, GnuPG rates it critical, no reported exploitation, update to GPGME 2.3.1.

Status: GPGME (GnuPG Made Easy) is the C library that many programs use to drive GnuPG. On November 20, 2026, GnuPG published an advisory for CVE-2026-129975, a double free in the library that it calls a "critical security bug found in all prior versions." GnuPG says a crafted OpenPGP key with a very long user ID can trigger it, that it "can be used exfiltrate data from applications," and that a remote code execution attack "can't be ruled out." The fix is GPGME 2.3.1, shipped in Gpg4win 5.1.2 for Windows. As of 02:31 EST on November 22, 2026, no source we found reports exploitation, and we found no public proof of concept. CVE-2026-129975 is not in CISA's Known Exploited Vulnerabilities (KEV) catalog. The CVE record, NVD and EPSS had no data for the ID when we checked, so the only severity ratings are GnuPG's "critical" and SUSE's own "important" (CVSS 3.1 8.1). Debian has fixed packages for bookworm, trixie and sid. Ubuntu and Red Hat pages were not available, and SUSE lists its packages as affected with the state Pending.

Executive Summary

GPGME sits between applications and the gpg program. Mail clients, file managers, signature checkers and server software call it to import keys, list keys and verify signed data. The flaw is in two internal functions that read output from gpg, so any program that lets GPGME handle an OpenPGP key from someone else is in scope. That includes a key attached to a message, fetched from a key server or uploaded by a user.

What matters for planning:

  • GnuPG's rating is critical, and the impact language is guarded. GnuPG says an attacker can use the double free to exfiltrate data from applications, with "sending for example signed data" as the example route, and that exploiting it for remote code execution "can't be ruled out." That is not a statement that code execution works. No source we found demonstrates it.
  • No exploitation is reported. GnuPG's announcement does not report any, and we found no other source that does. CISA has not listed the CVE in KEV. In Rescana's reading, the lack of a report is not proof of safety: the advisory was published on November 20, 2026, and the announcement is detailed about the root cause.
  • The affected set is large and hard to see. GnuPG's examples are Gpg4win, GnuPG Desktop, GnuPG VS-Desktop, "Most mail clients," "Many XMPP clients," Samba and Flatpak. Programs that link GPGME statically, or carry their own copy, need a rebuild, not a package update.
  • Patch coverage is uneven today. Debian shipped fixes for bookworm and trixie. SUSE shows Pending. We found no Ubuntu or Red Hat record. Debian's forky release still showed as vulnerable at 02:31 EST on November 22, 2026.
  • One easy-to-miss copy. The Ruby gpgme gem, version 2.0.26 (the latest we saw), ships GPGME 2.0.0 and builds it by default, so a system package update does not reach it unless the gem was installed with the system-library option.

If you run Gpg4win, update to 5.1.2. On Linux, install your distribution's update when it appears, restart anything that loaded the library, and then hunt for the copies a package update will not fix (Rescana guidance built on GnuPG's published steps). This is a new post: no live Rescana post covers CVE-2026-129975 or GPGME.

Technical Information

The following is GnuPG's published description. This post gives no payload details and links to no exploit code.

GnuPG's advisory (tracked as T8557, written by Werner Koch) says that "By using an OpenPGP key containing a crafted very long user id it is possible to trigger a double free." GnuPG attributes the discovery to Johan Carlsson and says GitLab informed GnuPG of it on November 12, 2026.

Root cause, per GnuPG. Two internal functions read status lines and key listings from gpg. A buffer of 1024 bytes is allocated at first and reallocated when more space is needed. The function keeps using the new pointer but only stores it back in the context object at the end. If the function returns early because of "an read error or other events," the context keeps a stale pointer to freed memory. When the context is released, that memory is freed a second time. GnuPG adds that "Due to other allocations in between the common double free checks of memory allocators may not be triggered." In Rescana's reading, that last sentence means the allocator's own double free abort is not a reliable alarm, so a clean log is weak evidence.

What the ticket adds. The GnuPG ticket T8557 quotes the original report, which names the functions read_status() and read_colon_line() in src/engine-gpg.c and says the early returns include cases where a status callback or the colon preprocess callback returns an error. The quoted report says the bug is present "also in 1.18.0 and later releases shipped in distributions." The fix commit message calls it "a 24 years old bug." Debian's tracker lists the introducing change as part of gpgme 0.3.9. The fix itself changes one source file: the buffer pointer is stored in the context immediately after the reallocation.

Severity, from whom. GnuPG calls it a "critical security bug." The CVE record did not exist at the CVE API, and NVD returned no entry, when we checked, so there is no CNA score and no NVD score. SUSE publishes its own CVSS 3.1 score of 8.1 (CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:H/I:H/A:H) and rates the issue "important," with the overall state Pending. Debian lists the urgency as "not yet assigned." These are different views: GnuPG's word reflects its judgment of impact, and SUSE's vector reflects a high attack complexity. We do not resolve the difference. The Rescana severity field for this post follows GnuPG's rating and attributes it.

Reach. GnuPG says all software using GPGME "prior to 2.3.0" is affected, "either as shared libary, DLL, or statically linked," and also says 2.3.1 fixes a bug "found in all prior versions." GnuPG notes that "version 2.3.0 was never publicly available," so for practical purposes the fixed release is 2.3.1. GnuPG says there is no need to update GPGMEPP or GPGMEQT.

Affected Product Versions

Product or component Affected Fixed or status Source
GPGME (libgpgme), all uses: shared library, DLL or statically linked All versions before 2.3.1 ("prior to 2.3.0" in the Impact section; 2.3.0 was never public) 2.3.1 GnuPG announcement
GPGME 1.24 branch Not stated by GnuPG The GPGME git mirror shows the same fix followed by a "Release 1.24.4" commit (05:22 EST, November 20), and GnuPG's FTP server lists a gpgme-1.24.4 tarball; GnuPG's announcement does not name 1.24.4; the next commit on that branch declares it "has reached end-of-usefullness" (06:32 EST, November 20); in Rescana's reading, no further 1.24 fixes should be expected GPGME git mirror, GnuPG FTP listing
Gpg4win Versions before 5.1.2 5.1.2, released November 20, 2026, with GPGME 2.3.1 Gpg4win change history
GnuPG Desktop, GnuPG VS-Desktop Named by GnuPG as users of GPGME GnuPG says new versions were released; no version numbers given GnuPG announcement
Debian bookworm 1.18.0-3, trixie 1.24.2-3, forky 2.2.0-2 bookworm 1.18.0-3+deb12u1 (DLA-4924-1), trixie 1.24.2-3+deb13u1 (DSA-6646-1), sid 2.3.1-1; forky still vulnerable when checked Debian security tracker
SUSE and openSUSE gpgme listed Affected on current and LTSS SUSE Linux Enterprise, SUSE Linux Micro and openSUSE Leap 16 products, and on older end-of-life products State Pending; SUSE Bugzilla 1291147; no SUSE security announcement linked SUSE CVE page
Ubuntu No CVE page found Unknown Ubuntu security site (404 when checked)
Red Hat No Red Hat record found (see the note below on RHEL 9) Unknown Red Hat Customer Portal data API (404) and Red Hat Bugzilla (no bug with the CVE alias)
Self-managed GitLab GitLab states no affected range for this entry GitLab's changelog lists "Validate OpenPGP input before passing it to GPGME" in 19.6.1, 19.5.8 and 19.4.10; that is an input check, not a GPGME update, and the entries do not cite the CVE GitLab changelog
Ruby gpgme gem 2.0.26 Ships GPGME 2.0.0 and builds it unless installed with --use-system-libraries No gem release with GPGME 2.3.1 seen on RubyGems RubyGems, gem contents
Samba and others named by GnuPG Named only as examples of software that uses GPGME No Samba advisory found GnuPG announcement, Samba security page

Three points on this table, all Rescana's reading. First, the fixed version for a distribution is that distribution's package version, not 2.3.1: Debian's bookworm and trixie fixes are backports onto 1.18.0 and 1.24.2. Second, we tie the GitLab changelog entries to this bug by their wording, their timing (GnuPG says GitLab informed it and moved the disclosure date up for GitLab) and the comment in the omnibus-gitlab source, which on its master branch builds GPGME 2.2.0 with a local patch that cites T8557. GitLab's patch release page for November 20, 2026 does not mention GPGME, and we could not confirm which GitLab packages ship that patch. Third, on RHEL 9: a message forwarded to the oss-security list says Red Hat Enterprise Linux 9 carried a downstream change for the same code "5 years ago," and the CentOS Stream 9 package source we opened does contain a 2021 patch, titled "src/engine-gpg.c: Fix memleaks," that stores the buffer pointer right after the reallocation in both functions. We did not test an RHEL 9 build and Red Hat has published nothing we could find, so do not treat RHEL 9 as safe on that basis. The same message says the patch is "included in RHEL 9 but not previous or subsequent releases," which, if accurate, means other Red Hat releases and Fedora lack it.

Timeline (EST)

GnuPG's announcement, ticket and the mailing list posts carry their own zones (CET, UTC, UTC-8 or UTC-5). Everything below is converted to EST.

  • November 12, 2026. GnuPG says GitLab informed it of the bug (date only).
  • November 12, 2026, 21:17 EST. GnuPG's ticket T8557 is created, quoting the report.
  • November 16, 2026, 11:42 EST. The ticket status moves from Open to Testing.
  • November 18, 2026, 04:21 EST. A GnuPG maintainer notes the fix is in GPGME 2.3.1, that a tarball and patch went to the reporter for distribution to Linux distributions, and that the announcement was planned for the 25th.
  • November 19, 2026. GPGME 2.3.1 is dated in its release notes. The release commit in the git mirror is at 12:05 EST (git timestamps are set by the committer and are not proof of when a change became public).
  • November 20, 2026, 05:16 to 05:22 EST. The git mirror shows the fix and a "Release 1.24.4" commit on the 1.24 branch.
  • November 20, 2026, 13:01 EST. GnuPG's announcement goes out on gnupg-announce. It says the date moved up from November 25 because GitLab had to update early for another bug it calls Ruby2Shell. Gpg4win 5.1.2 is released the same day (date only).
  • November 20, 2026, 15:56 EST. The announcement is forwarded to the oss-security list.
  • November 20, 2026, 20:06 EST. On the gnupg-devel list, one participant says the patch did not make it into the announcement mail (time not shown in the forwarded text) and another replies at this time pointing to the fix commit and to a 2021 Red Hat patch. That reply is forwarded to oss-security at 05:23 EST on November 21.
  • November 22, 2026, 02:26 to 02:33 EST. We checked the sources listed below. State at that time: no CVE record at the CVE API, NVD had no entry, EPSS had no data, not in KEV (catalog 2026.11.20), Debian fixed in bookworm, trixie and sid, SUSE Pending, no Ubuntu or Red Hat page.

The raw times in the Debian, SUSE, GitHub, RubyGems and FTP data are UTC where they appear. All dates and times in this post are EST.

How to Find Your Exposure

Evidence first. GnuPG publishes no indicators of compromise, and its own description says the allocator may not catch the double free, so exposure is about inventory, not log hunting (Rescana guidance, except where noted).

  1. Windows endpoints. Check the installed Gpg4win version. Anything before 5.1.2 carries a vulnerable GPGME. Search the file system for libgpgme DLLs outside the Gpg4win folder rather than relying on installed-program lists: other Windows applications and portable GnuPG builds may bundle their own, and GnuPG notes that a DLL can be taken from the updated Gpg4win.
  2. Linux and other Unix hosts. List installed packages built from gpgme (RPM distributions) or gpgme1.0 (Debian source name) and compare them with the fixed versions in the table above. Then find programs that load the library, for example with ldd against installed binaries or by looking for libgpgme in the memory maps of running processes. A running process keeps the old library until it restarts.
  3. Static and bundled copies. GnuPG says statically linked binaries must be rebuilt against 2.3.1. Ask each software vendor which GPGME version their product embeds. Check language bindings and wrappers that embed or link the core library (the Python gpg module, Perl and Go wrappers, Qt and KDE applications are Rescana's examples; GnuPG says only that GPGMEPP and GPGMEQT need no update, which does not mean the core library underneath them is fixed). For Ruby, gem list gpgme and the gem's ports/archives directory show whether the vendored 2.0.0 copy is present.
  4. Flatpak and container images. GnuPG lists Flatpak. Check application and runtime versions. Rebuild container images from patched base layers, and scan images for libgpgme shared objects or statically linked copies, because package scanners miss vendored ones.
  5. Where untrusted keys arrive. Make a list of services that import or list OpenPGP keys from outside parties, such as code hosting platforms that verify signed commits, mail gateways and mail clients, file-transfer or package tooling that verifies signatures, and chat clients. Those are the places where a crafted key can reach the library. GnuPG's description covers key listings as well as status lines, so a crafted key already sitting in a keyring may matter, not only one imported today. This is Rescana's prioritization and reading, not a GnuPG instruction.
  6. Crash evidence, with a caveat. Look for unexplained crashes or memory-corruption messages in processes that use GPGME, and for keys with unusually long user IDs in imported keyrings. GnuPG gives no length threshold beyond saying the read buffer starts at 1024 bytes, and a double free can pass silently, so an empty result does not clear a host.

Workaround and Mitigation

GnuPG's published remediation, in its wording: "Please update your GPGME shared library to version 2.3.1. If you need a DLL you may take it from from the updated Gpg4win version." For statically linked binaries: "remmeber to build them against GPGME 2.3.1." If GPGME cannot be updated: "you may patch your GPGME version using the attached patch which likley applies cleanly even to very old versions." GnuPG also says: "There is no need to udpate GPGMEPP or GPGMEQT." (Quoted as published, typos included.)

What that means in practice (Rescana guidance):

  • Windows: install Gpg4win 5.1.2 and confirm the version afterward. Verify the installer against the checksum on the Gpg4win download page (listed under Indicators of Compromise as an integrity value).
  • Debian: apply DSA-6646-1 on trixie and DLA-4924-1 on bookworm. Debian's tracker shows forky as vulnerable, so check for an update there.
  • SUSE, Ubuntu, Red Hat and others: no fixed package was published when we checked. Watch each vendor's tracker for CVE-2026-129975 and install the update when it appears. If you build your own packages, GnuPG's fix is the small change in the commit referenced from Debian's tracker.
  • After any update: restart every service and application that loaded libgpgme, because a running process keeps the old copy in memory.
  • Ruby gem users: a system GPGME update does not reach the vendored copy. The gem's own installer message says that installing with gem install gpgme -- --use-system-libraries links against the system library instead, and the gem's build script also reads the environment variable RUBY_GPGME_USE_SYSTEM_LIBRARIES; do either only after the system package is patched, and rebuild bundled deployments (Rescana guidance).
  • GitLab self-managed: versions 19.6.1, 19.5.8 and 19.4.10 carry GitLab's input validation change for GPGME input (changelog entry). GitLab's patch release page strongly recommends that all self-managed installations upgrade to one of these versions immediately; the only fix it lists is a separate Ruby issue (CVE-2026-132825), and the page does not mention GPGME. In Rescana's reading, upgrading is the practical step.
  • If you cannot update soon: limit the paths that feed untrusted OpenPGP keys to GPGME until the fix is in place, for example by turning off automatic key import from unknown senders. GnuPG does not list this as a workaround; it is Rescana guidance and we have not tested it against this bug.

Indicators of Compromise

GnuPG publishes no indicators of compromise for CVE-2026-129975: no hashes of malicious keys, no network addresses, no process names and no actor names. We do not publish any, and we found no public proof of concept or exploit code, so there is nothing to match against.

Two values are published for checking downloads. They are integrity checksums for the fixed software, not signs that a system was attacked:

Item Published checksum Source
gpgme-2.3.1.tar.bz2 1134a407b50204ee6fb9efaa6e1451c2460e543e (SHA-1, as printed by GnuPG) GnuPG announcement
gpg4win-5.1.2.exe 6bd36865f951c299e559ecb891f22e19b6fa0f4dfedc596f73d259f8365408eb (SHA-256) Gpg4win download page

GnuPG also points to its integrity check page and to signature files for each download. Prefer the signature check where you can. As Rescana guidance, treat unexplained crashes of GPGME-using programs after they process a key or signed message, and any attacker-supplied key with an unusually long user ID found in a keyring, as reasons to investigate that host. We will update this advisory if GnuPG or others publish exploitation details or indicators.

Third-Party Risk (TPRM) Angle

GPGME is rarely on a software bill of materials as a named product. It tends to arrive inside mail clients, backup and file-transfer tools, code hosting platforms, document tools and Windows desktop bundles, and in some products as a private copy that no package manager updates (Rescana's reading). That makes this advisory a supplier question as much as a patching job. Rescana has no data on how widely the flaw is used or exploited.

Suppliers to ask first (Rescana guidance):

  • Vendors of mail, secure messaging and signing products that import keys from outside parties on your users' machines or on gateways.
  • Hosting and developer-platform providers that verify signed commits or packages, including self-managed code hosting run by a managed service provider.
  • File-sharing and storage suppliers whose products GnuPG names (Samba) or that embed GPGME for signatures.
  • Suppliers that ship their own bundled GPGME, including Ruby applications that use the gpgme gem with its default build.

Questions that give a usable answer: Does any product we use embed or link GPGME, and which version? Do you process OpenPGP keys or signed data supplied by outside parties? Which release carries GPGME 2.3.1 or a backport of the fix, and when will it ship? Until then, what controls limit untrusted key import? Ask for dates, and ask whether statically linked or bundled copies are rebuilt as well as system packages.

Need help working out which suppliers use GPGME-based software with access to your data? Our team can help you prioritize them and track their answers. Talk to an expert

Sources

Talk to an expert

Talk to Rescana about this advisory — or about how we help TPRM teams keep reviews moving.