A malicious OpenPGP key with a bloated user ID can now pull data out of any application still linked against GPGME before 2.3.0.

GnuPG flagged CVE-2026-129975 this week as a critical memory bug in the library almost every program uses to talk to GnuPG. That is a cybersecurity problem hiding in mail clients, package signers, backup tools, and ticket systems that never show up on your internet-facing inventory. Fixed GPGME 2.3.0 is out. Your binaries may still be a week behind.

GPGME Trusted a Hostile User ID

GPGME is the glue. Apps call it so they don’t have to speak OpenPGP themselves. Import a key, verify a signature, decrypt a blob. Developers treat that as a solved primitive. Attackers treat it as a parser that runs inside your process.

The trigger is ugly in a useful way. An unusually long user ID on a crafted OpenPGP key can trip the memory bug and expose data from the host application. The exploit path is a key import. Someone in the org opens, refreshes, or trusts a file they believe is legitimate, often after a brute-force against the mailbox already failed.

OpenPGP padlock graphic illustrating the GPGME key-handling flaw
GnuPG’s GPGME library is how most applications import and use OpenPGP keys. A hostile user ID is enough to turn that parse into a data leak.

Sysadmins have spent years teaching people to verify fingerprints. That ritual does not save you if the parser dies before the check finishes. This is a bad look for every product that shipped PGP support as a checkbox and never pinned the library version underneath.

Your firewall never sees this. The key arrives over mail, a software mirror, a chat attachment, or an internal keyserver your threat-protection stack already trusts. Defense in depth that stops at the perimeter assumes the dangerous bytes are packets. Here the dangerous bytes are a contact card.

CISA’s Friday dump makes the same point with louder names.

Cybersecurity Patching Still Trails Live Exploits

The U.S. Cybersecurity and Infrastructure Security Agency added four flaws to its Known Exploited Vulnerabilities catalog, citing evidence of active use. Apache Roller, Cisco Expressway, SAP NetWeaver, and Broadcom VMware vCenter all landed on that list. CVE-2026-129448, a deserialization of untrusted data bug, carries a 9.8 CVSS score. The reporting around Friday’s drop already points to espionage use.

Roller is a blog engine. Expressway is a collaboration edge. NetWeaver sits under ERP. vCenter is the hypervisor console. Different logos, same class of mistake: software that accepts complex objects from people you semi-trust, then runs that parse on a host that can see everything. Your cyber security program still files those boxes under vendor patch Tuesday.

VMware and Cisco branding used in coverage of actively exploited infrastructure flaws
CISA’s KEV additions cover collaboration, ERP, blogging, and hypervisor consoles already under exploitation. The common trait is trusted parsers on privileged hosts.

KEV is a confession that threat detection arrived after the break-in.

Teams still wait for the named appliance advisory while GPGME sits in a dozen desktop apps with no owner. Incident response playbooks start at the SIEM. This class of bug starts at a library soname and a package that apt, winget, or a vendor installer quietly shipped two years ago.

Security hardening that only bumps the OS and the edge box leaves the crypto binding and the collaboration server as the live path. That is the thread connecting these stories. Hostile structured data gets parsed in a privileged process while your monitoring watches the wrong door.

Treat Crypto Bindings as Production Code

Stop treating GnuPG as a CLI hobby and GPGME as someone else’s problem. If an app can decrypt, sign, or import keys, it is production crypto, and it belongs on the same patch bar as your identity stack.

Immediate work is inventory, not another advisory meeting. Find every package and vendor binary still linked to GPGME before 2.3.0. Rebuild, upgrade, or isolate anything that still imports keys from mail, git, or internal keyservers. Block key import on jump boxes and admin workstations until the library is current.

For the KEV set, treat any internet-reachable Apache Roller, Expressway, NetWeaver, or vCenter host that missed the patch window as compromised until you prove otherwise. Rotate credentials those boxes could see. Pull process, deserialization, and session logs before you reboot the evidence away.

Shared libraries belong on the software bill you actually review. Pin GPGME and similar parsers in CI. Fail the build when a mail client or backup agent vendors a stale copy. Put key import, plugin upload, and gadget-heavy XML endpoints on the same allow-list discipline you already use for admin SSH. Log parse failures as security events, not application noise.

Give one owner the library list.

If your threat-protection stack cannot inspect an OpenPGP blob, constrain who can import keys and which hosts run GPGME at all. Then verify the soname. The splash screen on the vendor’s crypto feature is not a version number.

You’ll get another KEV drop. You will not get a second quiet week on GPGME if a mail client in finance is still on 2.2.x.

Sources

Take Control of Your Server Security

Don't let brute-force attacks slow you down. Try IPBan Pro risk-free for 30 days.

Secure. Automated. Lightweight.

Stay up to date with the latest news, releases and more.

Take Control of Your Server Security

Don't let brute-force attacks slow you down. Try IPBan Pro risk-free for 30 days.

Secure. Automated. Lightweight.