The UnrealIRCd project has warned that attackers are exploiting CVE-2026-114713, a flaw it describes as the most critical issue in the server’s 15-year history. “We are seeing bad guys exploiting this issue on the internet right now,” the project said in an exploitation notice added to its advisory.
How the Labeled-Response Module Lets Users Inject Server Commands
The vulnerability sits in UnrealIRCd’s labeled-response module, which handles the IRCv3 label tag. The module does not escape that tag when it echoes replies back. An ordinary user can therefore place server-to-server commands inside the label and have them run with server authority, including commands such as KILL and GLINE.
Exploitation requires a network with two or more linked servers. On such networks, a user with no operator privileges can issue actions normally reserved for the servers themselves, which means disconnecting other users, placing bans and carrying out other operator actions.
Severity Rating Gap Between the CVE Record and the Project
The CVE record rates the flaw CVSS 8.6, while the UnrealIRCd project calls it critical and places it above any previous issue in the software’s history. The difference reflects the project’s own view of the practical damage: a single unprivileged connection can alter the state of an entire linked network.
One of Nine Flaws Fixed in UnrealIRCd 6.2.7.1
The fix ships in UnrealIRCd 6.2.7.1, which resolves nine flaws in total, of which CVE-2026-114713 is one. The project also released a hot-patch that applies without restarting the server, an option that matters on IRC networks where a restart disconnects every user.
The project’s guidance sets out an order of operations. Administrators should first apply a spamfilter workaround, then move to the hot-patch or upgrade to 6.2.7.1. Operators still running UnrealIRCd 5 are told to blacklist the labeled-response module and migrate to version 6.
Internal Audit Aided by AI Found the Vulnerability
UnrealIRCd said it found the flaw during an internal audit “with the help of AI.” The project did not describe the tooling or process beyond that statement. The disclosure adds to a growing list of cases in which maintainers have credited AI assistance with surfacing serious bugs in long-lived code, although the project did not say whether attackers used similar methods to locate the issue.
Exploitation Timing and the Remaining Exposure
The project added the exploitation warning on November 9. It has not published the number of affected networks, the identity of the attackers or the activity they carried out after gaining server authority. The unanswered questions include whether the attacks so far have been disruptive, such as mass disconnections and bans, or something more targeted.
IRC networks with linked servers range from small hobby communities to older long-running public networks, and many run without dedicated security staff. The project’s warning that the flaw is being used in the wild means any such network that has not applied the spamfilter workaround, the hot-patch or the 6.2.7.1 upgrade remains open to any user who can connect. Because the only precondition beyond a linked-server setup is an ordinary user connection, no credentials or special access are needed to attempt the attack.
What Linked-Server Networks Stand to Lose
The consequences the project describes are operational. With server authority, an ordinary user can disconnect other users, place network-wide bans and run other operator actions, which in a community IRC network amounts to control over who can take part. The commands the project names, KILL and GLINE, are among the most disruptive an operator can issue, since the first removes a user from the network and the second bans a host mask.
The vulnerability is limited by one condition. A network running a single server does not expose the server-to-server command path, so the attack applies only where two or more servers are linked. Administrators running that layout are the ones the project’s advisory is aimed at.
The project’s position that this is the most serious issue in 15 years carries weight because the same advisory covers nine fixed flaws. Among those nine, the labeled-response bug is the one the maintainers chose to flag as under active attack.
