Back to Articles
high CVE-2026-114713

HIGH: UnrealIRCd Label Tag Flaw CVE-2026-114713 Under Active Attack

UnrealIRCd CVE-2026-114713 (CVSS 8.6) lets any user on a linked IRC network inject server-to-server commands through an unescaped IRCv3 label tag. Exploitation began a day after the November 7 hot-patch and CISA has added it to KEV. Upgrade to 6.2.7.1 now.

By Danny Mercer, CISSP — Lead Security Analyst • Nov 16, 2026
Is your business exposed? Our McKinney-based security team can assess your risk for free.
Share:

If you thought IRC was a quaint relic that only shows up in retro computing threads and the occasional botnet takedown, the last ten days should change your mind. UnrealIRCd, the most widely deployed open-source IRC server on the planet, just shipped a fix for what its own maintainer calls the most critical issue found in the project in the past 15 years. Attackers started poking at it roughly a day after the hot-patch landed, and CISA added it to the Known Exploited Vulnerabilities catalog on Friday, November 13, 2026. That is not a long runway.

The bug is tracked as CVE-2026-114713 and carries a CVSS score of 8.6. It lives in the labeled-response module of UnrealIRCd 6, and it affects every release from 6.0.0 up to but not including 6.2.7.1, which is the patched build. The project also released a hot-patch bundle on November 7, 2026 so administrators could close the hole without restarting their daemons. If you run an UnrealIRCd network and you have not touched it since early November, this is your sign to drop everything and patch it now.

What actually went wrong

IRCv3 introduced message tags, a way for clients and servers to attach metadata to individual messages. One of those tags is "label," which lets a client stamp a request with an identifier so it can match the server's reply to the command that triggered it. It is a genuinely useful feature for modern clients that pipeline commands. The labeled-response module takes whatever label the client sent and echoes it back in the reply.

The problem, according to the NVD description, is that the module does not escape the label tag when it echoes it into replies. Red Hat classifies the weakness as CWE-93, improper neutralization of CRLF sequences, and the general shape of the bug will be painfully familiar to anyone who has ever chased an HTTP response splitting issue. IRC is a line-based protocol. If an attacker can sneak line delimiters into a value that the server later writes back onto the wire, they can end the legitimate message early and start a brand new one of their own choosing.

On a single standalone server that is bad but somewhat contained. On a network with two or more linked servers it gets much worse, because those linked servers talk to each other using the server-to-server protocol and trust each other implicitly. An injected line that lands on a server link is treated as a command from a peer server, which means it executes with server authority. Per the NVD write-up, any user on such a network can exploit it. The Hacker News reporting describes the result as arbitrary command execution with server or IRCOp privileges, and Red Hat's summary notes that an unauthenticated network user could manipulate client tags to run commands with elevated server trust across linked daemons.

To be precise about what "command" means here, this is IRC protocol command injection, not a shell on the underlying Linux box. That distinction matters, but it is not as comforting as it sounds. Server-level authority on an IRC network means the attacker can do anything a linked server can do. Think forced nickname changes, kills, mode changes on any channel, spoofed messages that appear to come from services or trusted operators, bans against legitimate users, and general havoc that can take a community offline or turn its channels into a phishing platform. For networks that host development communities, open-source projects, or support channels, the trust abuse is the real payload.

Exploitation in the wild

The timeline here is the part that should worry defenders. UnrealIRCd released the hot-patch on November 7, 2026, and exploitation attempts reportedly began the very next day. That is classic patch diffing behavior. Somebody grabbed the fix, worked out what changed in the labeled-response module, and started firing crafted label tags at networks that had not updated yet. The specific goals of the attackers and the scale of the activity have not been disclosed, and nobody has published exposure counts for vulnerable servers.

CISA added CVE-2026-114713 to the KEV catalog on November 13, 2026, following reports of active exploitation, and gave Federal Civilian Executive Branch agencies until November 16, 2026 to remediate under Binding Operational Directive 26-04. A three-day deadline is CISA's way of saying the quiet part out loud. Most federal agencies are not running IRC networks, of course, but the KEV listing is a strong signal for everyone else that this is not theoretical.

Maintainer Bram Matthys, who goes by Syzop, did not mince words in his follow-up to administrators. "Hopefully everyone has hot-patched or updated their server, or at the very least has the spamfilter workaround deployed," he wrote. When the person who has shepherded a project for well over a decade starts using phrases like the most critical issue in 15 years, it is worth paying attention.

There is one more wrinkle worth noting. The 6.2.7.1 release fixed multiple vulnerabilities that were identified with the help of artificial intelligence, and this was one of them. That cuts both ways. The good news is that AI-assisted code review is finding real, high-impact bugs in mature C codebases before attackers do. The bad news is that the same tooling makes patch diffing faster for the other side, and the one-day gap between the fix and the first exploitation attempts is exactly what you would expect in that world.

Who should care

The obvious answer is anyone running UnrealIRCd 6 on a multi-server network. That includes the big public networks, but it also includes a surprising amount of infrastructure that nobody thinks of as IRC infrastructure. Plenty of organizations still run internal IRC for operations chat, on-call alerting, and build notifications because it is cheap, lightweight, and has worked fine for twenty years. Gaming communities, hobbyist networks, academic groups, and open-source projects lean on it heavily. Security teams themselves have a long tradition of running private IRC for incident coordination.

It is also worth remembering that IRC has a second life as malware command and control. Plenty of older botnets still use IRC channels to push commands to infected hosts. Defenders who operate sinkholes or research honeypots on UnrealIRCd should check their versions too, since a compromised research server is an awkward thing to explain.

Red Hat notes that it does not ship UnrealIRCd in RHEL or any of its other products, so a stock enterprise Linux install is not affected. The risk comes from servers that were built from source or installed from third-party packages, which is how almost everyone runs UnrealIRCd anyway. Do not assume your distribution's security feed will flag this for you.

What to do right now

The fix is straightforward. Upgrade to UnrealIRCd 6.2.7.1 or later on every server in the network, or apply the November 7 hot-patch if a restart is inconvenient. Because the attack abuses trust between linked servers, a partially patched network is still a problem. Every node needs the fix, including that forgotten leaf server somebody stood up in 2023 and never logged into again.

If you genuinely cannot patch today, deploy the spamfilter workaround that the UnrealIRCd team published alongside the advisory. Treat it as a stopgap and not a destination. Red Hat states that no mitigation it considers broadly applicable and stable is available, so the real answer is still the update.

Once you are patched, go looking for evidence that someone beat you to it. Review server logs from November 8 onward for client messages carrying unusually long or malformed label tags, and pay special attention to any label values containing carriage return or line feed characters. Check for unexpected server-originated actions such as kills, glines, mode changes, or nickname changes that no operator remembers issuing. Audit your operator list, services accounts, and channel access lists for anything that looks new. If your network relays messages to bots that trigger automation, such as deployment hooks or ticketing integrations, review those logs as well, because an attacker with server authority can spoof messages that those bots will happily trust.

Finally, take the opportunity to ask whether every linked server really needs to be linked, and whether any of them are reachable from the open internet when they do not need to be. Reducing the number of servers that can talk to each other reduces the blast radius the next time a bug like this shows up. And there will be a next time.

The bigger picture

There is a recurring lesson in this story that goes well beyond IRC. Echoing user input back into a protocol stream without escaping it is one of the oldest bug classes in the book. We have seen it with HTTP headers, SMTP, log files, and LDAP. Every time a protocol grows a new metadata field, somebody has to remember that the field is attacker-controlled. The labeled-response feature was added to make modern clients work better, and that is precisely why it deserved the same scrutiny as any other untrusted input.

The other lesson is about timing. The window between a patch shipping and exploitation starting has collapsed to roughly a day for projects with public source code. If your patch process for internet-facing services runs on a monthly cadence, you are effectively choosing to be exposed for most of that month. For anything on the KEV list, that cadence needs to be measured in hours.

If an IT provider manages your systems

Some businesses still run UnrealIRCd chat server software for internal chat, alerts or build notifications, often on a server nobody remembers setting up. Attackers began exploiting CVE-2026-114713 the day after the November 7, 2026 fix, and CISA lists it as actively exploited. Ask whoever runs your IT about it this week.

  • Do we run UnrealIRCd or any other IRC server anywhere, including old or test machines?
  • If so, is every linked server on version 6.2.7.1 or later, or hot-patched?
  • Have you reviewed the logs since November 8 for strange label tags or actions nobody approved?

A good answer is a confident "no, and here is how we checked," or proof that every server is updated and the logs were reviewed. If your provider cannot say for sure what software is running on your network, vulnerability scanning will find forgotten services like this one before attackers do.

If you are an IT provider yourself, see how we partner with MSPs.

References

Happening right now?

Business email taken over, or a payment sent to the wrong account?

Call us now. The first hours matter most: your bank may still be able to recall the payment, and whoever is in the mailbox keeps access until it is locked down. We handle the lockout and help trace the payment.

Not an emergency? Get a free security assessment instead.