CVE-2026-129428: Pulse read-only viewers could approve AI assistant commands

Pulse, a self-hosted dashboard for Proxmox, Docker and Kubernetes, let any signed-in viewer list, approve or deny the AI assistant's pending command approvals. Version 6.8.0 adds the missing permission check.

Christopher Ort
By Christopher Ort · 2 min read
Share
A server room with racks of equipment, a monitor showing a dashboard with pending approval requests, and a hand reaching toward the screen
Image: AI-generated, i10X
On this page
  1. Quick Take
  2. In plain English
  3. What happened
  4. What you should do
  5. What it means
  6. Sources

Pulse published advisory GHSA-9vxw-m79f-gxxr on 13 November 2026 for a missing authorization flaw reported by naroSEC. Versions 5.0.17 through 6.7.0 let an authenticated user holding only viewer access in an organization list pending AI Assistant approval requests and approve or deny them. VulnCheck assigned CVE-2026-129428 the next day.

i10X Quick Take

Quick Take

  1. What: Pulse versions 5.0.17 through 6.7.0 let any signed-in viewer approve or deny AI assistant commands without checking approval permission
  2. Who is affected: Self-hosted Pulse installations with the AI Assistant in use and more than one user in an organization
  3. What to do: Upgrade to Pulse v6.8.0, published on 13 November 2026
  4. Workaround: Grant viewer access only to users who should be able to see and decide pending approvals
  5. Fixed in: Version 6.8.0, first fixed prerelease v6.8.0-rc.2

In plain English

Pulse is a dashboard that lets you monitor servers and containers. It has an AI assistant that can run commands, but those commands need approval first. The bug let anyone who could log in and view the dashboard approve or deny those commands, even if they were only supposed to look, not decide. This means a viewer could see what commands the AI wanted to run and say yes or no, changing what actually happens and the record of who approved it.

What you should do: If you run Pulse with the AI assistant turned on and have multiple users, you should upgrade to version 6.8.0. If you don’t use Pulse, you can ignore this.

What happened

The approval routes in Pulse checked authentication, API token scopes, organization membership and request state, but never checked whether the signed-in user held approval permission. API tokens remained subject to their own scope checks, so the gap reached browser sessions.

The fix in pull request 4127 requires the same approval permission already used for the Actions review before a session can list, approve or deny a request.

The advisory states that this path does not by itself establish cross-organization access or command execution on a host. A viewer could read the proposed commands and target details in other users’ pending requests, and could approve or deny unexpired pending requests, which changes the approval decision and the audit record.

What you should do

Upgrade to Pulse v6.8.0, published on 13 November 2026. Until the upgrade lands, the maintainer’s workaround is to grant viewer access only to users who should be able to see and decide pending approvals.

To tell whether your Pulse instance was affected, compare the running version against the advisory’s range: v5.0.17 through v6.7.0 and the v6.8.0 prereleases beta.1, beta.2 and rc.1 are affected, and v6.8.0-rc.2 is the first fixed prerelease. The workaround points at the viewer role, so also review which users hold viewer access in each organization.

What it means

Pulse is self-hosted monitoring for Proxmox VE, PBS and PMG, Docker, Kubernetes, TrueNAS and vSphere, and it also exposes an MCP server for coding agents. Installations on v5.0.17 through v6.7.0, plus the v6.8.0-beta.1, beta.2 and rc.1 prereleases, with the AI Assistant in use and more than one user in an organization are affected.

The GitHub advisory assigned a CVSS score of 5.4 and classified the weakness as CWE-862, missing authorization.

Sources

Christopher Ort

Christopher Ort

Christopher Ort handles SEO, AI agent workflows and PR & partnerships at i10x.ai, the all-in-one AI workspace built by i10X Global Pte. Ltd. in Singapore. He is a graduate of Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU). He is the author of “The Style Penalty”, a study published on SSRN that analyses 1,576 evaluation decisions by four frontier AI models to show how AI-written CVs fare in automated hiring screens; the research has been covered by heise online and t3n.

Editor's picks

  1. AI-Powered Antidetect Browsers Enable Mass Social Media Fraud and Influence Operations
  2. Aokigahara Infostealer Targets AI Tools and Developer Credentials
  3. Yoba malware campaign abuses signed ChatGPT and Claude binaries via DLL sideloading
  4. Prompt Adware in Haiou Jailbreak Bundle Injects Ads into Codex Agent Responses
  5. ARTEX AI Hacking Platform Used in Cybercrime Campaigns