One UnrealIRCd Bug Lets Any Chat User Command the Whole Network
CVE-2026-114713 lets any user on a linked UnrealIRCd 6 network inject commands that run with server authority; exploitation began the day after the Nov. 7 fix and CISA has added it to KEV.
A bug in UnrealIRCd, the most widely deployed IRC server software, lets any ordinary chat user give orders with the authority of the server itself, and attackers started using it the day after the fix came out. The Hacker News reports that CISA added the flaw, CVE-2026-114713, to its Known Exploited Vulnerabilities catalog on Nov. 13, after UnrealIRCd’s maintainer Bram Matthys (Syzop) warned that “bad guys” were exploiting it. He calls it the most critical UnrealIRCd issue in 15 years.
For defenders
The bug: the labeled-response module in UnrealIRCd 6.0.0 up to 6.2.7 echoes the IRCv3 “label” message tag back into replies without escaping it. On a network with two or more linked servers, a normal user can use that to smuggle server-to-server commands that run with server authority (CWE-140, improper neutralization of delimiters). Per the 6.2.7.1 release notes, that means any user can KILL, GLINE “and do many more bad things”. A single-server setup isn’t exposed to this one, per the NVD description.
Status: CVSS 8.6 (NIST NVD, scope changed, integrity impact high). Fixed in 6.2.7.1, released Nov. 7, 2026 at 15:00 UTC along with a no-restart hot-patch for UnrealIRCd 6 on Linux and BSD (Windows needs the upgrade). Exploitation began Nov. 8, but what attackers did with it hasn’t been published: T1190 (Exploit Public-Facing Application). Public exploit code is reported in three GitHub repositories. Federal agencies must patch by Nov. 16. The same release fixes eight other bugs, including two ways to reveal a cloaked user’s IP address, all found in an AI-assisted audit.
Check yourself: run ./unrealircd version on every linked server; anything from 6.0.0 to 6.2.7 without the hot-patch is vulnerable. Then review server logs since Nov. 7 for KILL, GLINE or similar actions that no operator issued. No IOCs have been published.
What it means for you
If you don’t run an IRC server, there’s nothing for you to do. If you chat on IRC, a hijacked network could kick or ban users, or reveal people’s real IP addresses through the related bugs, so be patient with your network’s admins while they patch. If you run an UnrealIRCd network, this is a drop-everything fix: the hot-patch takes about a minute and doesn’t disconnect anyone.
Do this now
- Apply the hot-patch or upgrade to UnrealIRCd 6.2.7.1 or later on every server in the network, not just the hub.
- If you can’t patch today, deploy the spamfilter workaround the maintainer published on the UnrealIRCd forums.
- Review logs from Nov. 7 onward for kills, bans or mode changes nobody on your oper team made.
- Still on UnrealIRCd 5 or older? Those versions are unsupported; plan the move to 6.
Source article: Attackers Exploit UnrealIRCd Flaw That Enables Server-Level Command Injection
Publisher: The Hacker News (author: Ravie Lakshmanan)
Published: Nov. 14, 2026
URL: https://thehackernews.com/2026/11/attackers-exploit-unrealircd-flaw-that.html
Executive Summary
CVE-2026-114713 is a critical flaw in the labeled-response module of UnrealIRCd 6.0.0 through 6.2.7, the open-source IRC server maintained by Bram Matthys (Syzop). The module echoes the IRCv3 “label” message tag into replies without escaping it, so on any network with two or more linked servers an ordinary, unprivileged client can inject server-to-server commands that the receiving server executes with server authority; per the vendor this lets any user KILL, GLINE and perform other operator-level actions. UnrealIRCd 6.2.7.1 and a no-restart hot-patch shipped on Nov. 7, 2026 at 15:00 UTC, the maintainer reported in-the-wild exploitation starting the next day, and CISA added the flaw to its Known Exploited Vulnerabilities catalog on Nov. 13 with a Nov. 16 federal deadline. No actor has been named, and the attackers’ objectives, infrastructure and payloads have not been published; the sources describe the bug class (CWE-140, improper neutralization of delimiters), the affected versions, the fix and a spamfilter workaround, but no indicators, hashes or detection rules. The hunt leads below therefore focus on exposure, on the observable effect of exploitation (privileged actions with no operator behind them), and on the malformed label tags that carry the injection, with false positive expectations for each.
ATT&CK chain: Initial Access → Privilege Escalation → Impact
Hunt Lead Grid
| Hunt Lead Level | Hunt Lead Name | ATT&CK | Risk of False Positives |
|---|---|---|---|
| 1 (Junior) | Vulnerable Linked UnrealIRCd Servers Sweep | T1190 (Exploit Public-Facing Application) | Low. The lead anchors on exact version strings (6.0.0 to 6.2.7) on servers that have at least one server link; the only false positive is a server that was hot-patched without upgrading, which the hot-patch status check resolves. |
| 2 (Pro) | Privileged Actions With No Operator Behind Them | T1190 (Exploit Public-Facing Application); impact actions: not mapped (insufficient detail) | Medium. Services packages and automatic server bans issue kills and bans legitimately; the lead anchors on actions attributed to a server or user that is neither a known services server, an automatic ban source nor a logged-in IRCOp. |
| 3 (Elite) | Label Tags Carrying Unexpected Delimiters | T1190 (Exploit Public-Facing Application) | Medium. Some clients and bots generate long or unusual labels; the lead anchors on label values containing characters that act as delimiters in the IRC protocol, correlated with a server-originated command seconds later. |
Vulnerable Linked UnrealIRCd Servers Sweep
The flaw affects UnrealIRCd 6.0.0 up to and including 6.2.7 and is only exploitable when the server is part of a network of two or more linked servers, because the injected text is executed as a server-to-server command. The vendor fixed it in 6.2.7.1 on Nov. 7, 2026 and shipped the same fix as a hot-patch, and exploitation began the following day. The detection logic is an inventory sweep: for every host running UnrealIRCd, record the version (./unrealircd version, the server’s VERSION reply, or the version string in the startup log), whether the server has any active link blocks, and whether the security hot-patch has been applied; any linked server on 6.0.0 to 6.2.7 without the hot-patch is exposed and was exposed from Nov. 8 onward.
Use configuration management or package inventory, external attack surface scans of IRC ports (6667, 6697 and any custom listeners), and the servers’ own startup logs. Scope every server in the network, not only the hub, since any linked server accepting user connections is a possible entry point. Do not rely on blocking IRC traffic at the perimeter as the fix: the service is public by design and the vendor’s hot-patch is the control. A vulnerable server that was online after Nov. 8 should be treated as possibly exploited and reviewed with the next two leads; this sweep generalizes to any future UnrealIRCd advisory because the hot-patch mechanism and version reporting are the same.
Hunt Lead Level: 1 (Junior Analyst)
ATT&CK: Initial Access → T1190 (Exploit Public-Facing Application)
Sources: “Attackers Exploit UnrealIRCd Flaw That Enables Server-Level Command Injection”, The Hacker News, Nov. 14, 2026, https://thehackernews.com/2026/11/attackers-exploit-unrealircd-flaw-that.html (source article); “UnrealIRCd release notes (6.2.7.1)”, UnrealIRCd project on GitHub, https://raw.githubusercontent.com/unrealircd/unrealircd/unreal60_dev/doc/RELEASE-NOTES.md (additional source)
Privileged Actions With No Operator Behind Them
Per the vendor, successful exploitation lets “any ordinary user” execute commands with server or IRCOp authority, explicitly including KILL and GLINE; because the injected command is processed as if it came from a linked server, the resulting action is logged with server authority rather than against an operator who authenticated with OPER. The public sources do not say what attackers did after exploiting it. The detection logic is to pull every KILL, GLINE, ZLINE, KLINE, SHUN, SAJOIN, SAPART, SVSNICK, mode change on channels or users, and spamfilter or ban list change logged since Nov. 7, 2026, and flag any whose source is not a known services server, not an automatic ban source, and not an IRCOp with a matching OPER login in the same session; flag especially any such action that follows within seconds of an ordinary user’s command on a different linked server.
Use the ircd log files or UnrealIRCd’s JSON logging, services logs (Anope or Atheme) to separate legitimate services actions, and the oper-login events that UnrealIRCd records. Sweep back to Nov. 7 on every linked server and keep the window open for 90 days, since a G-line or spamfilter added by an attacker persists in the ban database. Allowlist by the exact services server name and by documented automatic ban sources rather than by action type. Tuning out “noisy” server-sourced kills would be the wrong move here, because server-sourced authority is exactly what the bug forges. A confirmed unexplained privileged action means the network was exploited: patch every server, review and remove bans and spamfilters added in the window, and rotate oper and link passwords.
Hunt Lead Level: 2 (Pro)
ATT&CK: Initial Access → T1190 (Exploit Public-Facing Application); Impact → not mapped (insufficient detail on attacker actions)
Sources: The Hacker News (above), source article; UnrealIRCd release notes (above), additional source
Label Tags Carrying Unexpected Delimiters
The root cause, per the NVD description, is that the labeled-response module does not escape the client-supplied IRCv3 “label” tag when it copies it into replies, and the CWE classification is improper neutralization of delimiters. Legitimate clients use labels as short opaque tokens to match replies to requests, so a label that contains characters with structural meaning in the IRC line or message-tag format is the carrier for the injection. The detection logic is to capture or log client-sent message tags and flag any label value that contains spaces, semicolons, carriage returns or line feeds, or escape sequences, or that is far longer than the client’s other labels, and to correlate each such label with any server-originated command observed on a linked server within a few seconds of it. The maintainer also published a spamfilter workaround on the UnrealIRCd forums; the forum post could not be retrieved for this report, so its exact rule text is not reproduced here.
This needs packet capture on the plaintext or TLS-terminated client listener, or UnrealIRCd debug-level logging of raw client input, joined with server-to-server traffic logs on the links, which makes it a correlation exercise for a detection engineer. Baseline label formats per client software for a week before alerting, because some bots generate long labels. Stripping labeled-response support entirely is a heavier mitigation than needed once the hot-patch is applied, but it is a reasonable emergency stopgap on servers that cannot be patched immediately. Any malformed label followed by a server-authority command should be escalated as exploitation, and the pattern generalizes to other tag-echo injection bugs in IRCv3 implementations.
Hunt Lead Level: 3 (Elite)
ATT&CK: Initial Access → T1190 (Exploit Public-Facing Application)
Sources: The Hacker News (above), source article; “CVE-2026-114713 Detail”, NIST National Vulnerability Database, https://nvd.nist.gov/vuln/detail/cve-2026-114713 (additional source)
Hash Appendix
No malware, tools or files are associated with this exploitation in the source article or the additional sources, so there are no hashes to report. OTX AlienVault and VirusTotal were not queried because there is no file to pivot from.
References Appendix
- “Attackers Exploit UnrealIRCd Flaw That Enables Server-Level Command Injection”, The Hacker News (Ravie Lakshmanan), Nov. 14, 2026, https://thehackernews.com/2026/11/attackers-exploit-unrealircd-flaw-that.html (source article)
- “CISA Adds Two Known Exploited Vulnerabilities to Catalog”, CISA, Nov. 13, 2026, https://www.cisa.gov/news-events/alerts/2026/11/13/cisa-adds-two-known-exploited-vulnerabilities-catalog
- “CVE-2026-114713 Detail”, NIST National Vulnerability Database, https://nvd.nist.gov/vuln/detail/cve-2026-114713
- “UnrealIRCd release notes (6.2.7.1)”, UnrealIRCd project, https://raw.githubusercontent.com/unrealircd/unrealircd/unreal60_dev/doc/RELEASE-NOTES.md
- MITRE ATT&CK Enterprise technique pages linked above, MITRE, https://attack.mitre.org/
CVE at a glance
- CVSS
- 7.7 High (v3.1)
- EPSS
- 0.5% chance of exploitation in the next 30 days (higher than 43.8% of CVEs)
- Exploited in the wild
- Yes, on CISA’s KEV list since Nov 13, 2026
- Public exploit
- Yes, reported in 3 GitHub repositories
Data from NVD, FIRST EPSS and CISA KEV, refreshed automatically. We report whether public exploit code exists but never link to it: fake proof-of-concepts that infect researchers are common.