CRITICAL: Perforce Puppet CVE-2026-115731 RCE Exploited in the Wild
CISA has added CVE-2026-115731, a CVSS 9.4 deserialization flaw in Puppet Enterprise and Puppet Core report processing, to its KEV catalog after attackers began exploiting it a day after the November 3 patch. Any CA-signed agent certificate, or an autosign-enabled CA, is enough to run code on the Puppet Server and steal its signing key.
If you've ever wondered what the most dangerous server in your environment is, here's a hint. It's not the domain controller everybody obsesses over. It's the box that quietly tells every other box what software to install, which config files to write, and which commands to run as root on the next check-in. For a lot of enterprises that box runs Puppet, and as of this week it's on CISA's list of things attackers are actively breaking into.
On Tuesday, November 17, CISA added CVE-2026-115731 to its Known Exploited Vulnerabilities catalog. The flaw is a deserialization of untrusted data bug in the report processing code of Perforce Puppet Enterprise and Puppet Core, and it gives an attacker remote code execution on the Puppet Server as the service user. Perforce rates it 9.4 on the CVSS scale, and some trackers score it even higher under CVSS 3.1 scoring. Federal civilian agencies have until December 1 to remediate under Binding Operational Directive 26-04. Everybody else should treat that date as generous.
What actually broke
The bug lives in how Puppet Server rebuilds the event data inside an agent's report. Every agent sends a report back to the server after a run, and that report travels to the PUT /puppet/v3/report endpoint on the HTTPS port, TCP 8140 by default. According to a technical breakdown from Ostorlab, the method Puppet::Transaction::Event#initialize_from_hash handed that event data to Puppet's rich data deserializer, the FromDataConverter. That converter trusts a reserved key called __ptype to tell it what kind of object to reconstruct. An attacker who supplies an Object type carrying a RubyMethod annotation gets that type turned into a real Ruby class by the RubyGenerator, and the embedded Ruby then runs on the server through class_eval. In other words, the server politely compiles and executes whatever the client describes. It's the classic deserialization story, just wearing a Puppet costume.
The fix swaps that path over to the ToStringifiedConverter, which reads the event hash as plain strings and never parses types at all. Ostorlab notes that other deserialization paths still exist in the codebase, which is a good reason to install the vendor release rather than trying to hot patch a single Ruby method yourself.
There is one gate in front of all this. The request has to carry an agent certificate signed by the Puppet certificate authority. That sounds reassuring until you remember how many shops run Puppet with autosign turned on so new nodes can enroll without a human clicking approve. CISA's own description spells it out plainly. If the CA signs requests without strong verification, an unauthenticated attacker on the network can simply ask for a certificate, receive one, and walk straight through the front door. And if autosign is off, any single compromised agent anywhere in the fleet still holds a valid certificate, which turns one owned workstation or web server into a path to the system that manages all of them.
Why this one hurts more than the usual RCE
Code execution on a random application server is bad. Code execution on the configuration management server is a different category of bad. Perforce itself warns that a successful attack could expose the server's certificate signing key. Once an attacker holds that key, they can mint trusted certificates for any hostname they like and change the software or configuration that gets delivered to managed systems on their next Puppet run.
The researchers who found the bug, Alain Rödel of Neodyme AG, put it bluntly. Regardless of configuration, an attacker with RCE on a Puppet server can obtain the CA signing keys and impersonate other hosts, and because Puppet typically has broad reach across a corporate fleet, that access lends itself to ransomware deployment, mass data exfiltration, and lateral movement. Picture a ransomware crew that doesn't need to push its payload host by host because your own orchestration platform will happily do it for them on a thirty minute schedule. That's the threat model here.
Rödel also reported a companion bug, CVE-2026-115732, rated 9.0. It's a certificate name authorization bypass affecting both Puppet and OpenVox, where a certificate whose name extends an approved name can slip past authorization rules and impersonate a managed machine. It isn't on the KEV list yet, but it was patched in the same releases and you should treat the pair as a single job. Neodyme said the team found both issues with AI assistance and careful steering, which is a sign of where vulnerability research is heading whether we like it or not.
Exploitation in the wild
Perforce shipped fixes on November 3. By November 4, a Reddit user posting as transdimensia reported that the flaw was already being exploited. They described finding a severely misconfigured CA in their own fleet and tracing an intrusion back through CVE-2026-115731 itself. A proof of concept targeting OpenVox, the community fork of Puppet, was published shortly after the patch landed, so the barrier to entry is now roughly the effort it takes to copy and paste.
There are no public details yet on who is behind the attacks or what payloads they're dropping. That's not unusual this early, but it does mean you can't rely on a neat set of indicators of compromise to tell you whether you're clean. Absence of evidence is going to require some actual looking.
Affected versions and patches
On the commercial side, Puppet Enterprise and Puppet Enterprise Advanced 2026.0.0 are fixed in 2026.0.1, which ships Puppet Core 9.1.3. The 2025 stream from 2025.0.0 through 2025.12.0 is fixed in 2025.12.1, and the long term support 2023.8 stream from 2023.8.0 through 2023.8.12 is fixed in 2023.8.13, both bundling Puppet Core 8.22.2 under the hood. Standalone Puppet Core is fixed in 9.1.3 for the 9.x line, covering 9.0.0 through 9.1.2, and in 8.22.2 for the 8.x line, covering everything from 8.0.0 through 8.22.1.
OpenVox users need to update two pieces. OpenVox Agent is fixed in 9.3.0 and 8.32.0, while OpenVox Server is fixed in 9.2.0 and 8.18.0. The relevant GitHub advisories are GHSA-ffr6-rwpm-p5wq for CVE-2026-115731 and GHSA-q4hm-33rc-572g for CVE-2026-115732 in trapperkeeper-authorization. Restart the server after its bundled Puppet library is updated, because a patched gem sitting on disk next to an unpatched process in memory protects nobody.
What to do right now
Patch first. It's the only complete fix, and the release paths above are straightforward upgrades within each supported stream. If you're one of the organizations still running a Puppet Enterprise version older than 2023.8, this is the week that technical debt comes due.
While the change window gets scheduled, close the unauthenticated path. Run puppet config print autosign on the primary server and look at autosign.conf. If it says true or contains a broad wildcard, turn it off or replace it with policy based autosigning that actually verifies who is asking. Then restrict TCP 8140 so only your managed agents and consoles can reach it. A Puppet Server listening to the entire internet, or to an entire flat corporate network, is doing nobody any favors. Be clear eyed about what these steps buy you, though. They remove the drive by angle, but a stolen or legitimately issued agent certificate still reaches the vulnerable code.
Then go hunting, because exploitation started a day after disclosure and you may already be late. Run puppetserver ca list --all and look for certificate names that don't match anything you manage. Search the Puppet Server access logs for PUT requests to /puppet/v3/report coming from certificate names that aren't your agents, or from unusual source addresses. Review processes spawned by the puppet service user and any recently modified files in its reach. The public proof of concept drops a /tmp/pwned marker, which is worth checking for, but real attackers are rarely that courteous, so don't treat a missing file as a clean bill of health. It's also worth diffing your control repository and recent catalog changes against what your team actually committed, since an attacker with server access is likely to tamper with what gets pushed downstream.
If you find an unknown certificate or any sign of code execution, assume the CA is compromised. That means rebuilding the server, rotating the CA, and reissuing certificates to every node. It's painful, and it's the only answer that holds up, because a stolen signing key doesn't stop working just because you patched the bug that leaked it.
The bigger picture
Infrastructure as code was supposed to make environments consistent and auditable, and it does. It also concentrates an enormous amount of trust in a handful of servers that many teams set up years ago, tuned for convenience, and haven't looked at closely since. Autosign was a reasonable shortcut when someone was enrolling the first hundred nodes. Today it's the difference between an authenticated bug and an unauthenticated one. If your configuration management, RMM, or deployment tooling hasn't had a hardening review in a while, CVE-2026-115731 is a pretty loud reminder to give it one.
MSP angle
Plenty of mid market clients have a Puppet, Ansible, or similar automation server that nobody on staff fully owns anymore, which makes a fixed price "management plane hardening" assessment an easy conversation to start this week. Pair the patch and autosign review with a compromise assessment and ongoing log monitoring for the orchestration tier, and you've turned an emergency fix into a recurring managed detection line item.
References
- The Hacker News, Hackers Exploiting Critical Puppet RCE Flaw
https://thehackernews.com/2026/11/hackers-exploiting-critical-puppet-rce.html
- SecurityWeek, Critical Perforce Puppet RCE Exploited in Attacks
https://www.securityweek.com/critical-perforce-puppet-rce-exploited-in-attacks/
- Ostorlab, Puppet Server Deserialization RCE (CVE-2026-115731)
https://blog.ostorlab.co/puppet-server-deserialization-rce-cve-2026-115731.html
- Red Hat CVE-2026-115731
https://access.redhat.com/security/cve/CVE-2026-115731
- NVD CVE-2026-115731
https://nvd.nist.gov/vuln/detail/CVE-2026-115731
- CISA Known Exploited Vulnerabilities Catalog
https://www.cisa.gov/known-exploited-vulnerabilities-catalog
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.
Protect Your Organization
Find vulnerabilities like this in your systems before attackers do.
24/7 monitoring to detect and respond to threats like these in real time.
Block phishing and malware delivery targeting your organization.
Map security controls to 26 frameworks including NIST, SOC 2, and HIPAA.