CVE-2026-77087: Paperclip DNS rebinding remote code execution

CVE-2026-77087 · Severity: critical · CVSS 9.6 · Published 2026-08-21

Paperclip is a development framework that allows developers to build AI agents locally. By default, it runs in a trusted mode that automatically grants full administrative access to requests from localhost. An attacker can exploit DNS rebinding—a technique where a domain name resolves to both an attacker's server and the victim's local machine—to trick a developer's browser into making authenticated requests to their local Paperclip instance. This allows the attacker to execute arbitrary commands on the developer's computer, potentially stealing source code, credentials, SSH keys, and installing backdoors, all without the developer's knowledge or interaction beyond visiting a malicious webpage.

Quick answers

Is CVE-2026-77087 exploited in the wild?
No. Junglewise Threat Intelligence has no report of CVE-2026-77087 being exploited in the wild.
How severe is CVE-2026-77087?
Critical severity, with a CVSS score of 9.6, published 21 August 2026.
Which products and versions are affected?
Paperclip Paperclip before 0.3.1.
Is there a fix for CVE-2026-77087?
No fixed version is recorded yet. Check the vendor's advisory in the references below. If you run it, Junglewise drafts a mitigation that fits your setup and applies it once you approve, with senior security engineers behind it.
How is CVE-2026-77087 attacked?
It is attacked over the network, in a software library.
Am I affected by CVE-2026-77087?
Check whether anything you run is named in the advisory. Junglewise does that check for you around the clock: connect your code, cloud and domains, and it matches new vulnerabilities like this one against what you actually run and tells you when one reaches you.

Technical details

The vulnerability is a missing Host header validation (CWE-862) in Paperclip's local_trusted deployment mode. Three factors combine: (1) local_trusted mode auto-authenticates all requests as instance admin with no checks; (2) Host header validation is disabled in local_trusted mode, only active in authenticated+private mode; (3) the process adapter executes arbitrary commands via spawn() without sandboxing. An attacker registers a domain with two DNS A records—one pointing to their server, one to 127.0.0.1 (localhost)—and crafts a malicious webpage. When a victim visits it, the browser connects to the attacker's server, which serves JavaScript that then makes API calls. As the server shuts down, the browser falls back to the 127.0.0.1 address, hitting the local Paperclip instance. Since Host header validation is missing, Paperclip accepts the request, auto-grants admin access, and the JavaScript creates a company with a process adapter configured to execute arbitrary bash commands. The attack requires only that a developer visit the malicious page; no credentials, clicks, or additional interaction are needed. No patch is currently available beyond upgrading to 0.3.1 or later.

Affected products

  • Paperclip Paperclip before 0.3.1

Timeline

  • 2026-07-22: disclosed: GitHub security advisory GHSA-x8hx-rhr2-9rf7 published
  • 2026-08-21: advisory: CVE-2026-77087 published on NVD

References

This page as Markdown. Free to reuse under CC BY 4.0 with a link back.