Saturday, November 21, 2026
News

Is Puppet's 9.4 flaw exploited, and what must you patch?

Three of the first reports on this flaw say it is being exploited, no attacker path has been published, and the vendor says it is not aware of attacks. The patch list and the log line to grep for are not in dispute.

Two blue cabinets filled with complex electronic components

Puppet Enterprise 2026.0.1, 2025.12.1 and 2023.8.13 carry the fix for CVE-2026-115731, as do Puppet Core 9.1.3 and 8.22.2. CISA listed the flaw in its Known Exploited Vulnerabilities catalog on 17 November and set a 1 December deadline, though Perforce says it sees no attacks.

Which Puppet versions are affected, and what is fixed?

The CVE record names three Puppet Enterprise ranges and nothing wider. They are 2023.8.0 to 2023.8.12, 2025.0.0 to 2025.12.0, and 2026.0.0 on its own, which makes the fixed releases 2023.8.13, 2025.12.1 and 2026.0.1. The record marks those three exact releases unaffected, so the bound excludes them. Security Affairs lists the same three. Scoping the ticket to those ranges is the same exercise as pinning down which Java versions Java4Shell touches.

Puppet Core is in scope too, and the record's own configuration list is not where you will find it. The Hacker News names Puppet Core 9.0.0 before 9.1.3 and Puppet Core 8.x before 8.22.2. Ostorlab puts the ranges more exactly at 9.0.0 to 9.1.2 and 8.0.0 to 8.22.1. SecurityWeek says Puppet Core 9.1.3 and 8.22.2 patch both flaws, the second being CVE-2026-115732, which Rescana scores 9.0 and says is not in the KEV catalog. For the OpenVox fork, The Hacker News gives fixed releases of Agent 9.3.0 and 8.32.0 and Server 9.2.0 and 8.18.0.

Upgrading is not the whole step. The fix text on the record requires a restart before the patched code is in use, so a package upgrade left running on the old process is still exposed.

Is CVE-2026-115731 actually being exploited?

The sources that broke this story do not agree, and that is worth knowing before you write the ticket. SecurityWeek reports that attackers "have been exploiting" the flaw, and that it "was flagged as exploited on November 4". Security Affairs says the opposite in plain words, that "no reports confirming active exploitation" exist. Perforce, it adds, is "not aware of attacks" exploiting it, and no threat actor and no campaign has been tied to it.

The Hacker News sits between the two. It reports the KEV listing, which by definition records exploitation, then says no details have been published on how the flaw is being weaponized in the wild. The 4 November flag traces back to a Reddit post, from a user who described finding "a severely misconfigured CA" in their own fleet. Treat that thread carefully: the URL The Hacker News cites for it names CVE-2026-115732, the companion flaw, not this one.

What is not in dispute is the listing and the clock. The KEV data on the CVE record gives a date added of 17 November 2026 and a due date of 1 December 2026, which Ostorlab and Rescana both repeat, and which leaves federal agencies 10 days from today. The supporting signal has also moved. Rescana, checking on 19 November, found CISA's exploitation value for the flaw reading "active". On the CVE record it flipped from none to active on 18 November and back to none on 20 November, and it reads none now. Nobody has explained the second flip. The gap between a catalog entry and the published evidence is the same one that ran through the poisoned agent-framework releases.

What can an attacker do after it works?

A Puppet server holds the certificate authority for the fleet, which is what makes this more than one host. Neodyme AG, whose researcher Alain Rödel reported the flaw, says that once code runs on the Puppet server, an attacker can take the CA signing keys and then pose as any other host. Code runs as the Puppet Server service user, and from there the configuration that the server pushes out is also in reach.

The mechanism is a report the server trusts too much. Ostorlab's write-up traces it to the report event hash being read with FromDataConverter, which rebuilds object types from an attacker-controlled __ptype value, and says the fix moves to ToStringifiedConverter. The record files it as CWE-502, deserialization of untrusted data, and CWE-94, code injection. It scores 9.4 on CVSS 4.0, on the vector CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:H/VI:H/VA:H/SC:H/SI:H/SA:H, and 9.9 on NVD's CVSS 3.1. Both vectors assume the attacker already has low privileges, which in practice means a signed agent certificate, which is why autosign keeps coming up.

How do you tell whether your Puppet server was hit?

There are no published indicators, so this is log work rather than a hash list. Rescana says that no indicators of compromise have been published by Perforce, Belgium's CCB or CISA, and Ostorlab says nothing public exists "beyond the proof-of-concept marker", that marker being the public exploit's touch /tmp/pwned. A separate proof-of-concept aimed at OpenVox appeared, per SecurityWeek, "shortly after CVE-2026-115731 was patched". Ostorlab and Rescana both put working exploit code in public on 3 November, the day the fix shipped, which is sooner than SecurityWeek's wording suggests.

Rescana publishes the checks it would run, as its own guidance rather than vendor instructions. For the version: cat /opt/puppetlabs/server/pe_version, cat /opt/puppetlabs/puppet/VERSION, and rpm -q puppet-agent or dpkg-query -W puppet-agent. Ostorlab adds puppet --version, rpm -q puppetserver puppet-agent and dpkg -l puppetserver puppet-agent. For the attack path, Rescana points at the access log /var/log/puppetlabs/puppetserver/puppetserver-access.log and the request PUT /puppet/v3/report/, and at the directory from puppet config print reportdir --section server for a __ptype value beginning Object. Check your exposure too: puppet config print autosign --section server, the file /etc/puppetlabs/puppet/autosign.conf, and /opt/puppetlabs/bin/puppetserver ca list --all from Ostorlab. The service listens on TCP 8140 by default. Rescana adds TCP 8142 for the orchestrator broker, citing the reporter's write-up.

What should IT teams do about Puppet CVE-2026-115731 today?

  • Upgrade to 2026.0.1, 2025.12.1 or 2023.8.13 on Puppet Enterprise, or 9.1.3 or 8.22.2 on Puppet Core, then restart Puppet Server. The record says the restart is what loads the fixed code.
  • Treat 1 December as the date even if you are not a federal agency. That is 10 days out, and it is the only published deadline attached to this flaw.
  • Switch autosigning off, or make the autosign script validate requests strictly. That is the mitigation Rescana attributes to Perforce, and SecurityWeek and Security Affairs give the same advice. Rescana warns it does not stop anyone already holding a valid certificate.
  • Pull the access log for PUT /puppet/v3/report/ requests back to early November before you patch, since the patch closes the hole but tells you nothing about what came through it.
  • Restrict TCP 8140, the default port, to the hosts that need it, per Ostorlab, and check whether 8142 is reachable from anywhere it should not be.
  • Plan for the CA, not just the host. If you find evidence of code execution, the signing keys are in scope, on Neodyme's account of what the flaw gives up.
  • Patch CVE-2026-115732 in the same window. The Puppet Core releases cover both, and only one of the two is in the KEV catalog.

Sources