On November 17, 2026, CISA added CVE-2026-115731 to its Known Exploited Vulnerabilities (KEV) catalog for Perforce Puppet Enterprise. The flaw is deserialization of untrusted data in Puppet Server report processing: a client that holds any CA-signed agent certificate can send a PUT report containing a crafted rich-data type and run arbitrary code as the Puppet Server service user. Perforce’s November 2026 security releases ship fixes in 2023.8.13, 2025.12.1, and 2026.0.1. If your fleet still runs an affected build, patch the server first, then review CA autosign—CISA and GitHub advisory GHSA-4pxx-8frq-9r4j both warn that weak autosign can let an unauthenticated network party obtain a certificate and walk the same path.
This is not an AI-model bug; it is configuration-management infrastructure that many engineering orgs share with CI runners, cloud images, and developer laptops enrolled as Puppet agents. The KEV listing matters because attackers are already using it somewhere in the wild, even though public write-ups are still thin.
TL;DR
| Question | Answer |
|---|---|
| Am I affected? | Puppet Enterprise in the listed ranges without 2023.8.13 / 2025.12.1 / 2026.0.1 |
| Is it exploited? | Yes — CISA KEV date added 2026-11-17 |
| Is there a fix? | Upgrade Puppet Server / PE to a patched release; Puppet Core 9.1.3 and 8.22.1 also fix the same CVE pair |
| Worst case? | Full compromise of the Puppet Server host at service-user privilege, then malicious manifests for every managed node |
| Autosign angle? | Tighten or disable autosign if you cannot prove strong verification |
| Due date (federal)? | 2026-12-01 per CISA catalog |
| Coding agents? | Agents on enrolled laptops are clients; risk is server takeover and bad catalog data, not the LLM itself |
What CISA and Perforce published
The KEV entry names Perforce | Puppet Enterprise, cites CWE-502 (deserialization) and CWE-94 (code generation), and points to Perforce’s report-deserialization advisory URL and NVD CVE-2026-115731. Forensic triage under BOD 26-04 is listed No for this CVE in the catalog snapshot Beam reviewed on November 17, 2026; ransomware campaign use is Unknown.
Perforce’s whats-new page groups CVE-2026-115731 with CVE-2026-115732 (certificate name authorization bypass) in the same November 2026 maintenance wave. The vendor describes 115731 as closing arbitrary code execution on the Puppet Server host and 115732 as stopping one managed machine from partially matching another machine’s certificate name to impersonate it.
The GitHub Advisory Database entry repeats the technical sentence used across NVD and CISA: deserialization in report-processing code, exploitation via crafted rich-data type in a report submitted through PUT, prerequisite of a CA-signed agent certificate, with the autosign escalation called out explicitly.
Attack chain step by step
Understanding the chain helps you decide whether monitoring agent laptops is enough (it is not) versus patching the server (it is mandatory).
Step 1 — Attacker holds or obtains an agent identity
Puppet agents authenticate with certificates signed by the deployment’s Puppet CA. The vulnerability is not “anonymous RCE on port 8140”; it is RCE as an authenticated agent (or anything that can present a valid agent cert).
If autosign is enabled without strong checks (DNS names only, predictable patterns, open lab networks), the advisory text treated as authoritative by CISA says an unauthenticated network attacker can request and receive a signed certificate. That turns a deserialization bug with low privilege required in CVSS terms into a pre-authentication foothold in poorly configured fleets.
Step 2 — Crafted report hits report processing
After enrollment, agents periodically submit reports to Puppet Server summarizing last run state. The server deserializes rich structured fields into internal objects. The flaw is unsafe deserialization: attacker-controlled content in a rich-data type inside the report becomes executable logic on the server during processing—not merely bad logging.
Delivery is a normal PUT on the report API surface agents already use. No novel port or admin UI path is required once the certificate exists.
Step 3 — Code runs as Puppet Server service user
Successful exploitation executes as the Unix service account running Puppet Server (commonly pe-puppet or equivalent in PE layouts). That user can read signing keys, catalog secrets, Hiera backends reachable from the server, and often pivot to other management integrations wired into PE (console DB, webhooks, cloud provisioners).
Step 4 — Downstream impact on managed nodes
Puppet Server is the source of truth for configuration. A server compromise is a supply-chain pivot: catalogs and facts can be altered so every agent pulls attacker-controlled manifests on the next run. That pattern is adjacent to how teams worry about compromised agent skills on workstations, but here the blast radius is fleet-wide OS and package state, not a single IDE session.
Affected and fixed versions
Perforce’s public release notes and the GitHub advisory agree on version bands:
| Track | Affected | Fixed |
|---|---|---|
| 2023.8.x | ≥ 2023.8.0, < 2023.8.13 | 2023.8.13 |
| 2025.x | ≥ 2025.0.0, < 2025.12.1 | 2025.12.1 |
| 2026.0.x | ≥ 2026.0.0, < 2026.0.1 | 2026.0.1 |
Open-source Puppet Core releases 9.1.3 and 8.22.1 carry the same two CVE fixes for teams not on PE entitlements but running Puppet Server from core packages.
CVE-2026-115732 ships in the same maintenance wave
Perforce paired CVE-2026-115731 with CVE-2026-115732 (certificate name authorization bypass). That bug let a managed node partially match another machine's approved certificate name and receive configuration meant for a neighbor. You need both fixes from the November 2026 releases, not a cherry-picked JAR.
In incident response terms, 115731 is server-side code execution; 115732 is wrong-node identity that makes post-exploitation persistence harder to spot in compliance audits. Agent users rarely touch Puppet YAML directly, but platform engineers paste agent-generated config refactors into control repos—after server compromise, catalog diffs deserve the same review skepticism you apply to unexpected Claude Code settings edits.
Federal timing and BOD 26-04 context
CISA's catalog row sets Date Added 2026-11-17 and Due Date 2026-12-01. The entry links BOD 26-04 prioritization guidance and notes forensic triage is No for this specific CVE at listing time—meaning agencies focus on patch or discontinue, not a mandated incident hunt template from that flag alone.
Private companies often mirror KEV anyway because insurers and customer security questionnaires track the same dates. If your coding-agent fleet sits in a SOC2 scope that includes Puppet-managed build images, cite this KEV entry in this week's change ticket so patch windows beat holiday code freezes.
Patch, verify, and harden
Upgrade path
- Read Perforce’s release notes for your track on puppet.com whats-new.
- Schedule Puppet Server / PE console maintenance; agents can often stay enrolled while the server upgrades, but plan for catalog compile downtime.
- After upgrade, confirm the running version matches 2023.8.13, 2025.12.1, or 2026.0.1 (or newer) via your standard admin command or package query.
- Roll Puppet agents only if release notes require a minimum agent version for protocol compatibility—deserialization fix is server-side, but PE upgrades frequently bundle console and agent compatibility notes.
Autosign and certificate hygiene
Even on a patched server, stale autosign policy is how red teams turn “authenticated-only” bugs into internet-facing pre-auth chains:
- Disable autosign except tightly scoped break-glass workflows.
- Require manual approval or external signing integrations for new nodes.
- Inventory orphaned certificates issued during lab bursts; revoke if the node no longer exists.
- Mirror the mindset from GitGuardian-style credential hygiene: anything that auto-trusts new identities will eventually meet an RCE.
Signs you may have been targeted
Public sources do not ship IoCs for this CVE as of November 17, 2026. Reasonable server-side checks include:
- Unexpected report payloads or errors in Puppet Server logs around report submission times
- New agent certificates you did not approve
- Catalog changes not tied to approved commits in your control repo
- Process spawns from the Puppet Server JVM user unrelated to compile cycles
If you are federal or follow CISA timing, plan remediation before the December 1, 2026 KEV due date.
What this means if you run AI coding agents
Coding agents (Claude Code, Codex, Cursor cloud agents, custom MCP harnesses) do not patch Puppet for you. They do often run on the same laptops and CI hosts that are Puppet-managed:
- An agent with shell access on an enrolled laptop can read agent SSL material, local Hiera snippets, or
puppet.confthat names internal hosts—useful recon after any server breach. - Conversely, a compromised Puppet Server can push packages, sudoers, or cron jobs that alter the environment agents inherit on the next run, including installing untrusted tools or weakening firewall rules.
- Agent sessions that auto-apply infrastructure changes amplify impact: if your workflow lets an agent open merge requests and your runners are Puppet-enrolled, treat server integrity as part of agent supply-chain defense, similar to how teams responded to Chrome KEV issues on developer browsers.
Practical checks today:
- Patch Puppet Server before debating agent prompt policy.
- On developer machines, run agents under workspace policy that denies destructive
puppetresource commands if you do not intend agents to mutate enrollment. - Rotate any secrets that lived in Hiera data reachable from a server you have not yet proven is patched.
- Read CVE volume trends as context: KEV additions are accelerating; configuration-management RCE belongs in the same patch queue as language-runtime CVEs.
What Beam can and cannot do here
Beam watches coding-agent tool use on the workstation—credential patterns, destructive shell, MCP and skill heuristics—not Puppet Server JVM deserialization. It will not detect crafted rich-data reports on port 8140.
Where Beam still helps on enrolled laptops: PreToolUse deny (Claude Code and Codex) can block agent-driven curl exfiltration of /etc/puppetlabs/ material; skill and MCP scanning catches risky install scripts an attacker might try to plant after a catalog change. Those controls are heuristic and hook-based, not a substitute for PE patching.
Related reading
- Chrome CVE-2026-125350 on CISA KEV — another November 2026 KEV item on developer machines
- CVE volume in 2026 — prioritization when advisories spike
- Active Storage CVE-2026-122006 — deserialization-adjacent risk in app stacks agents modify
- What AI agent monitoring tracks — on-device action visibility vs server-side config management
Frequently asked questions
The FAQ in the front matter mirrors the questions above; short answers are repeated here for on-page readers.
What is CVE-2026-115731? Critical Puppet Server report deserialization RCE with CA-signed agent certificate prerequisite.
Is it exploited? CISA KEV listing on November 17, 2026.
What versions fix it? Puppet Enterprise 2023.8.13, 2025.12.1, 2026.0.1 (and Puppet Core 9.1.3 / 8.22.1).
Autosign risk? Weak autosign can let strangers obtain certificates; close that gap while patching.
Primary sources
- CISA Known Exploited Vulnerabilities Catalog — CVE-2026-115731
- GitHub Advisory GHSA-4pxx-8frq-9r4j / CVE-2026-115731
- Perforce Puppet Enterprise whats-new — November 2026 security releases
- CVE Record CVE-2026-115731
Version numbers, KEV metadata, and vendor release notes are accurate as of November 17, 2026; confirm current builds on puppet.com before maintenance windows.

