UnrealIRCd's CVE-2026-114713 Lets Any User Run Server Commands, and Exploitation Began a Day After the Nov. 7 Patch
Security / news
UnrealIRCd's CVE-2026-114713 Lets Any User Run Server Commands, and Exploitation Began a Day After the Nov. 7 Patch
CISA listed the label-tag injection on Nov. 13; versions 6.0.0 through 6.2.7 are affected and 6.2.7.1 or a no-restart hot-patch fixes it.
Any ordinary user connected to an UnrealIRCd network with two or more linked servers can inject server-to-server commands that run with server authority, if the network runs a version from 6.0.0 through 6.2.7. The flaw is CVE-2026-114713, CISA added it to its Known Exploited Vulnerabilities catalog on 13 November 2026, and the fix is version 6.2.7.1 or a hot-patch that needs no restart.
What the flaw lets a user do
Bram Matthys (known as Syzop), UnrealIRCd's maintainer, posted the advisory to the oss-security list on 8 November, a day after the 6.2.7.1 release and hot-patch on 7 November. The bug is an escaping failure in the IRCv3 "label" message tag. A user can put newlines into server-to-server traffic, which lets them issue commands "with server authority", including KILL and GLINE. Matthys calls it "the most critical issue found in UnrealIRCd in the past 15 years".
The Hacker News gives the NVD score as 8.6 under CVSS 3.1. The score understates the damage on a network that cares about its operator tier: no account, no operator status and no network access beyond a normal client connection is needed, only a multi-server deployment. A single-server network is not exposed to the server-to-server path the advisory describes.
Exploitation and the CISA listing
The Hacker News quotes Matthys: "We are seeing bad guys exploiting this issue on the internet right now". The Hacker News reports that exploitation began about a day after the hot-patch and that the attackers' goals are unknown. CISA added the entry on 13 November under the title "UnrealIRCd Improper Neutralization of Delimiters Vulnerability", citing active exploitation. The Hacker News gives a federal deadline of 16 November under BOD 26-04; the CISA page itself states none.
| Date (2026) | Event |
|---|---|
| 7 November | UnrealIRCd 6.2.7.1 and hot-patch bundle-20261107 released |
| 8 November | Advisory posted to oss-security |
| About 8 November | Exploitation begins, per The Hacker News |
| 13 November | CISA adds CVE-2026-114713 to KEV |
| 16 November | Federal deadline, per The Hacker News |
How to fix it, in the order the advisory gives
Step one is a stopgap. An IRC operator can add a spamfilter, SPAMFILTER add -regex T block - Invalid_label ^label=(?![^\r\n;\x20]*\z), which covers the critical bug only if every server runs 6.0.4 or newer. The other eight fixed issues still need the patch.
Step two is either route. On each server, ./unrealircd hot-patch bundle-20261107 applies the fix live and prints "Patch installed and server rehashed correctly. All should be good now!" Windows users should upgrade instead, since the advisory excludes them from the hot-patch. On Unix, ./unrealircd upgrade then a restart installs 6.2.7.1.
To check, run ./unrealircd module-status|grep labeled-response. A line reading 5.0-hot-patch-bundle-20261107 means the hot-patch took, 6.2.7.1 means the fixed release is running, and a bare 5.0 means the server is still vulnerable.
UnrealIRCd 5 and the other eight bugs
UnrealIRCd 5 has been end of life since 1 July 2023 and is similarly affected, with no fix coming. For it, the advisory says to add blacklist-module "labeled-response"; to the config on every server and restart rather than rehash. Matthys says several of the bugs likely reach back to version 3.2.x, which he did not test.
The same release fixes eight other issues:
- A user can reveal another user's IP address, plus a separate IP leak
- A blacklist recheck can hang the server
- A crash on CHGIDENT from Services that a normal user can trigger
- A counting error that makes a server answer "This server is full" to every new connection until restart
- Services can set at most twice
set::maxbansbans - A possible flood through TLS renegotiation
- A possible crash on whowasdb
Found with AI help, which is the pattern now
The advisory credits no outside researcher. It says the issues were found "during an internal audit, with the help of AI". That fits two stories on this site: Google's report on an exploit agent diffing patches and Project Zero's disclosure deadlines, which cut the window for researchers to hold a bug back. Here the reported gap between patch and exploitation was about a day.
Operators who have not run the module-status check should do it today. The remaining unknown is who is exploiting the flaw and against how many networks; neither the advisory nor CISA says.
Sources
More in Security
- 01Ruby2Shell (CVE-2026-132825): Ruby Is Fixed, but GitLab, GitHub Enterprise and Chatwoot Are Listed Without a PatchA one-byte heap overflow in Ruby's case conversion carries a 7.4 local score, while the researchers say stock Rails is remotely reachable and will publish a proof of concept Nov. 25.
- 02NPCI Wants Overlay Permissions Restricted on Sideloaded Apps; Google's 2024 India Block List Omits ThemThe reported ask covers accessibility, device admin and display-over-apps. The pilot Google announced on Oct. 3, 2024 named four other permissions.
- 03CVE-2026-99312: A KVM Nested-AMD Bug Scores 8.8, With Fixes Listed for Three Kernel BranchesThe Linux CNA's scoring notes describe a freed page-table root that a guest can aim at; the commit message describes only a stale MMU context.
- 04CISA's Nov. 18 KEV Additions Include Apache Axis 1.x, Which Has No Fix, With a Nov. 21 DeadlineThree of four new entries carry a three-day federal deadline; LiteLLM's advisory names two different patched versions.