If attackers reached your Puppet server through CVE-2026-115731, upgrading closes the hole but does not evict anyone. Remote code execution on the server puts the certificate authority (CA) signing key within reach, and with that key they can mint a valid identity for any machine in your fleet. The real fix is patch, then decide whether the CA can still be trusted.
CISA added the flaw to its Known Exploited Vulnerabilities (KEV) catalog on November 17, 2026, two weeks after Perforce and the OpenVox project shipped fixes on November 3. Federal civilian agencies have until December 1. Everyone else running Puppet to manage Linux servers should treat it as already late.
What CVE-2026-115731 lets an attacker do
CVE-2026-115731 is a critical (CVSS 9.4) deserialization flaw in how Puppet Server processes the run reports agents send back after each run. Any machine with an agent certificate signed by the Puppet CA can submit a crafted report and run code on the server as the Puppet service user. If the CA signs certificate requests automatically, the attacker does not need a certificate to begin with.
The researcher who found it, Alain Rödel of Neodyme, traced the bug to agent-supplied type annotations that the server turns into Ruby code and evaluates. Neodyme says the vulnerable path dates back at least to open source Puppet 6.24.0 from July 2021, and that it works against a default configuration. The report endpoint lives on port 8140, the same port every agent already talks to.
The same disclosure carried a sibling, CVE-2026-115732 (CVSS 9.0): a hostname check that compares only as many domain labels as the shorter name has, so a certificate for evil.web01.example.com can pass as web01.example.com. That bug is not in KEV, but it changes how you audit, as below.
Which versions are fixed
| Release line | Vulnerable | Fixed in |
|---|---|---|
| Puppet Enterprise 2026.0 | 2026.0.0 | 2026.0.1 |
| Puppet Enterprise 2025 | before 2025.12.1 | 2025.12.1 |
| Puppet Enterprise 2023.8 LTS | before 2023.8.13 | 2023.8.13 |
| Puppet Core 9 | before 9.1.3 | 9.1.3 |
| Puppet Core 8 | before 8.22.2 | 8.22.2 |
| OpenVox Server 8 / 9 | before 8.18.0 / 9.2.0 | 8.18.0 / 9.2.0 |
| Open source Puppet (legacy) | all known versions | no updates, move to OpenVox |
Puppet Enterprise is fixed in 2026.0.1, 2025.12.1 and the 2023.8.13 long-term release. Puppet Core is fixed in 9.1.3 and 8.22.2. On the community fork, the OpenVox advisory lists openvox-server 8.18.0 and 9.2.0, openvoxdb 8.18.0 and 9.2.0, and openvox-agent 8.32.0 and 9.3.0 as the minimums. The same advisory says open source Puppet will receive no further updates, so anyone still on those packages has no patch coming and should move to OpenVox or Puppet Core.
Is it exploited, and who says so?
Sources disagree, and the gap matters for your risk call. CISA lists the flaw as exploited. A Puppet administrator posted on Reddit on November 4, one day after the fixes, describing an intrusion they traced to a badly misconfigured CA; The Hacker News noted that the post's link names the sibling CVE-2026-115732 rather than this one. Security Affairs reported that, per the vendor's latest information, Perforce is not aware of attacks or customer data access.
Our read: the KEV listing is the operative signal. No threat actor or campaign has been named publicly, and no indicators have been published. That absence is a reason to check your own server, not a reason to wait.
Why patching alone does not restore trust
A configuration management server is a different kind of victim from a web app. It holds three things an intruder wants, and the patch protects none of them retroactively.
-
The CA private key. Neodyme notes that code execution on the server is enough to take it. A stolen key keeps working after the upgrade, because every agent still trusts that CA.
-
The catalog. On servers where the Puppet service user can write the code that gets compiled into catalogs, an attacker can push changes that every managed host applies at its next check-in. Neodyme calls the hardened setup, where a different user owns that code, the line that limits this.
-
Orchestration. Where Puppet Enterprise's PXP agent or Choria on OpenVox is enabled, Neodyme says the blast radius can extend to root on managed hosts.
So the question after patching is not "are we fixed" but "was the server touched before we fixed it". If you cannot rule that out, rotating the CA and re-enrolling agents is the only clean answer, painful as it is. A server that only patched still trusts every certificate the old key ever signed.
Autosign is the line between authenticated and not
On paper, exploitation needs a signed certificate. In practice, Neodyme writes that most large companies run autosign, often with a broad pattern such as *.example.com so provisioning tools can enroll new machines without a human. With a wildcard rule, anyone who can reach port 8140 and pick a matching name gets a certificate, which turns this into an unauthenticated bug.
The sibling flaw makes the wildcard worse. Under a pattern that accepts subdomains, an attacker can request a name one label deeper than a real host and, until CVE-2026-115732 is patched, use it to impersonate that host. When you audit, extra-deep names are the thing to look for.
Check your server before you close the ticket
Three checks cover most of the exposure. Turn off autosign until fixed builds are in place, or replace a wildcard with policy-based signing that verifies something the attacker cannot guess. List every certificate the CA has signed and match it against your inventory, flagging names with more labels than your naming scheme uses. Then look at who has been sending reports.
# /etc/puppetlabs/puppet/puppet.conf [server] autosign = false # every certificate the CA has signed puppetserver ca list --all # agent reports arrive as PUT requests on port 8140 grep 'PUT /puppet/v3/report/' /var/log/puppetlabs/puppetserver/puppetserver-access.log
A report from a certname that is not in your inventory, or one that appears only around early November, deserves a closer look. Agents report on every run (every 30 minutes by default), so a malicious report blends into routine traffic. The certificate inventory is a cleaner signal than the traffic itself.
If you cannot upgrade today, OpenVox suggests limiting which network segments can reach the server and tightening autosign. It also published standalone patch scripts, with the caveat that they may not apply cleanly to older builds. After upgrading, revalidate any custom auth.conf rules, since OpenVox warns some may have relied on the buggy matching.
A pattern worth naming
This is not the first Puppet issue this year to center on the management layer: in July we covered a Resource API bug that wrote hidden passwords in cleartext on every node. The same shape shows up in other control planes, from Rancher's plaintext cluster tokens to the N-able N-central RMM console. Whatever pushes configuration to every server deserves the network isolation and logging you would give a domain controller. If you want an outside pair of eyes on that kind of event, our incident response page describes how we approach it.