
Research writeup on CVE-2026-114713, an IRCv3 label tag injection in UnrealIRCd's labeled-response module enabling server-to-server command injection, with detection and remediation guidance.
CVE-2026-114713 — UnrealIRCd labeled-response IRCv3 label injection (server-to-server command injection)
| CVE | CVE-2026-114713 |
| Component | UnrealIRCd, module labeled-response (IRCv3 label message tag support) |
| Affected | UnrealIRCd 6.0.0 – 6.2.7 (and UnrealIRCd 5 EOL) |
| Fixed in | UnrealIRCd 6.2.7.1 · hot-patch bundle-20261107 |
| CWE | CWE-140 — Improper Neutralization of Delimiters |
| Impact | Server-to-server command injection with full server/IRCOp authority |
| Vector | Network, unauthenticated (any ordinary client on a multi-server network) |
| Trigger | Client-supplied IRCv3 label tag containing delimiter characters (\n, \r, etc.) that is echoed unsanitized into replies |
| CVSS 3.1 | 8.6 High (CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:C/C:N/I:H/A:N) |
| Status | Actively exploited · CISA KEV (added 2026-11-13, remediation due 2026-11-16) |
| Credit / discovery | Internal UnrealIRCd audit (assisted by AI); advisory by Bram Matthys |
| Advisory | https://seclists.org/oss-sec/2026/q4/576 · https://forums.unrealircd.org/viewtopic.php?t=9487 · https://nvd.nist.gov/vuln/detail/CVE-2026-114713 |
| Workaround | blacklist-module "labeled-response"; or network-wide SPAMFILTER |
The labeled-response module implements the IRCv3 label message tag. When a client includes a label tag on a request, the server is expected to echo the same label on the corresponding reply so the client can correlate responses. The module failed to escape or neutralize delimiter characters inside the tag value before writing it into the outbound message.
IRC is a strictly line-oriented protocol: messages are terminated by \r\n (or \n). Because the label value was inserted without sanitization, an attacker could embed newline characters (and other protocol delimiters) inside the tag. The resulting wire data terminated the current message early and caused the remainder of the “label” to be parsed as one or more independent server-to-server (S2S) commands.
On any network consisting of two or more linked servers the injected commands are processed with full server authority. An ordinary, unprivileged user can therefore issue commands such as KILL, GLINE, KLINE, TKLINE, server notices, or other privileged S2S operations that affect the entire linked network.
This is the most severe issue reported in UnrealIRCd in the last fifteen years. A single unauthenticated client connection is sufficient; no operator privileges or special modules beyond the default labeled-response load are required.
IRCv3 message tags are key-value pairs carried in the @ prefix of an IRC message. The label tag is a client-generated opaque string that the server must echo on the matching reply. The vulnerable code path in the labeled-response module performed a direct (or insufficiently escaped) insertion of the client-supplied value into the reply buffer.
Because the insertion did not neutralize the line terminators that IRC uses as message delimiters, the following sequence becomes possible:
Client connects to any server on a multi-server network and sends a command that supports labeled responses (e.g. PRIVMSG, PING, WHOIS, …) carrying a crafted tag:
@label=benign\r\nSERVER injected.example :arbitrary S2S payload
The module echoes the label into the reply without stripping or escaping \r / \n.
The receiving server (or any downstream linked server) parses the injected material as a new, independent S2S command and executes it with server privileges.
The attack surface exists only when the module is loaded and the network contains at least two linked servers. A single standalone UnrealIRCd instance is not vulnerable to the privilege-escalation aspect, although the same lack of escaping still constitutes incorrect protocol handling.
CVSS reflects a pure integrity impact with changed scope (network-wide effect from a single client). Confidentiality and availability are not directly scored, although secondary effects (mass disconnects, ban-list exhaustion, etc.) are possible.
| Series | Status |
|---|---|
| UnrealIRCd 6.0.0 – 6.2.7 | Vulnerable |
| UnrealIRCd 6.2.7.1 | Fixed |
| UnrealIRCd 5.x (EOL since 2023-07-01) | Also affected; no official fix beyond module blacklisting |
The vendor released both a full version bump (6.2.7.1) and a no-restart hot-patch (bundle-20261107) that can be applied to any 6.0.0–6.2.7 installation.
Preferred path (recommended order):
Immediate temporary mitigation (network-wide, if all servers ≥ 6.0.4)
As an IRCOp on any server:
SPAMFILTER add -regex T block - Invalid_label ^label=(?![^\r\n;\x20]*\z)
This blocks malicious labels before they can be echoed. It does not protect against the other eight issues fixed in 6.2.7.1.
*Hot-patch (no restart required on NIX)
./unrealircd hot-patch bundle-20261107
Verify:
./unrealircd module-status | grep labeled-response
Expected output for a successful hot-patch:
labeled-response 5.0-hot-patch-bundle-20261107
Or, after a full upgrade:
labeled-response 6.2.7.1
The same check can be performed on IRC with MODULE -all (and MODULE -all remote.server.name for remote servers).
Full upgrade
Upgrade every server to UnrealIRCd 6.2.7.1 (or later) and restart.
Permanent workaround (if patching is delayed):
blacklist-module "labeled-response";
Apply on every server and rehash/restart as appropriate. This disables the IRCv3 labeled-response feature entirely.
UnrealIRCd 5 operators should blacklist the module and migrate to a supported 6.x release as soon as possible.
MODULE -all or ./unrealircd module-status.label tag are blocked and logged.This repository is published by Sable Research for educational and defensive research purposes only.
The information contained herein is intended to help defenders understand and mitigate the vulnerability.
Do not use this information to attack systems you do not own or have explicit written permission to test.
Unauthorized exploitation of this vulnerability is illegal.