UnrealIRCd command injection exploited in the wild

Actively exploited in the wild - CVE-2026-114713 is a high-severity command injection flaw in UnrealIRCd 6.0.0 before 6.2.7.1 that lets any user on a multi-server network inject server-to-server commands executed with server authority. Patched in 6.2.7.1 - update immediately.

Overview

UnrealIRCd implements the IRCv3 labeled-response extension. When a client sends a message tag named label, the server is supposed to echo that value back in its replies so the client can correlate requests and responses. In affected versions, the label value is not escaped before it is copied into outgoing replies.

On a single standalone server this is harmless. On networks with two or more linked servers, the server-to-server protocol is a line-oriented text protocol. An attacker who controls the label content can embed line breaks and protocol tokens, so the crafted label is not echoed as data but interpreted as a fresh command issued by the server itself. Because those commands travel over the trusted server link, they run with server authority rather than the privileges of the originating user.

The practical consequence: a regular user with no special rights can cause a linked server to execute commands that should only ever come from an operator or a trusted peer. That can disrupt network state, manipulate channel and user metadata, desynchronize server links, or corrupt the trust relationship between linked nodes. The CVSS score of 8.6 reflects network reachability, low complexity, and no privileges or user interaction required. CISA has added this CVE to its Known Exploited Vulnerabilities catalog, and the EPSS score sits at roughly 0.3 percent for the next 30 days, so while exploitation is confirmed, mass scanning is not currently indicated.

Impact

• Any user on a network with two or more linked servers can trigger the flaw.
• Injected commands execute with server-to-server authority, not user authority.
• Networks can suffer link desynchronization, state corruption, or service disruption.
• Trust boundaries between linked servers are effectively bypassed for the attacker.

Single-server installations do not expose the server-to-server path and are not exploitable through this vector.

Remediation and Mitigation

1. Upgrade UnrealIRCd to 6.2.7.1 or later. This is the only complete fix.
2. Until every node is patched, keep the network topology in mind: an unpatched server anywhere on the mesh keeps the risk open, because the injection works through links.
3. If you cannot patch immediately, disconnect untrusted users from multi-server deployments or temporarily reduce to a single server.
4. After upgrading, review server logs for unexpected server-to-server command activity, link drops, or state changes around channel and user records.
5. Restrict who can register and connect during the exposure window, and audit operator and service accounts for unexpected changes.

For background on how exploited vulnerabilities are tracked and what data exposure looks like after an incident, see breach reports and current security news.

Security Insight

IRCv3 tag handling has become a recurring injection surface across IRC daemons: the same class of bug appeared in earlier tag-processing code paths where untrusted client values were concatenated into server messages without encoding. What stands out here is the trust model - the s2s protocol assumes every line from a linked peer is authoritative, so a single unescaped string collapses the distinction between user input and server command. Vendors building federated or mesh architectures should treat label and tag echo paths as equivalent to command parsers, because that is exactly how attackers will treat them.

Further Reading

• All High Vulnerabilities
• Recent Critical Vulnerabilities
• Top Critical CVEs