Loading vulnerability…
CVE-2026-41679 is a critical authorization bypass in Paperclip, a Node.js and React-based platform for orchestrating AI agents. Affected deployments are network-accessible instances running in authenticated mode with the default registration configuration prior to version 2026.416.0. The flaw stems from missing or insufficient authorization checks across the registration and import workflow, allowing an unauthenticated attacker to self-register without email verification, authenticate, create and approve a CLI authorization challenge, obtain a persistent board-level API credential, and then access a company import path that incorrectly accepted board-level privileges where instance-administrator privileges were required. By importing a crafted company bundle containing a .paperclip.yaml definition for an agent using the host-level process execution adapter, the attacker can cause Paperclip to launch an attacker-controlled command as a child process of the server. The issue is fully automatable, requires no user interaction, and can be exploited through a short API call chain against default vulnerable deployments.
Mallory correlates every CVE against your assets, your vendors, and active adversary campaigns. Know which vulnerabilities matter for you, not just which ones are loud.
What it means. What to do now. Patch path, mitigations, and the assume-compromise checklist.
What an attacker gets, and what they’ve been doing with it.
If you can’t patch tonight, do this now.
Patch, then assume compromise.
1 valid exploit after Mallory filtered fakes, detection scripts, and README-only repos.
This repository is a small Python proof-of-concept exploit for CVE-2026-41679 affecting Paperclip AI. The repo contains one executable script (poc.py), a short README with usage instructions, and a minimal requirements.txt listing requests. The exploit is not framework-based. The script performs a full exploitation chain against a target Paperclip instance over HTTP(S): it signs up a user, signs in to obtain a session cookie, creates a CLI auth challenge, approves that challenge with the authenticated session, then uses the returned board API token to call the company import API. In that import request it supplies inline files including COMPANY.md, agents/pwn/AGENTS.md, and a malicious .paperclip.yaml. The YAML defines an agent named pwn using a process adapter configured to run bash with '-c' and attacker-controlled commands. After import, the script calls the agent wakeup endpoint to trigger execution. The exploit’s main capability is authenticated remote command execution through abuse of the import and agent execution workflow. The default payload writes the output of 'id' and 'whoami' to /tmp/pwned.txt, but the -c argument allows arbitrary shell commands, making this more than a pure detector. TLS certificate verification is explicitly disabled, so it will also work against instances with self-signed certificates. Overall, this is an operational PoC that automates exploitation end-to-end and demonstrates code execution on the target host.
Products and vendors Mallory has correlated with this vulnerability. Open in Mallory to drill down to specific CPE configurations and version ranges.
Vendor-confirmed product mapping. Mallory continuously reconciles this list against your asset inventory.
17 sources tracked across advisories, community write-ups, and news. New activity surfaces here as Mallory finds it.
A critical authorization bypass in the Paperclip AI management platform that could let a remote attacker self-register, gain board-level API access, and achieve arbitrary code execution with the permissions of the Paperclip server process.
A critical Paperclip server-side command execution vulnerability in which a newly registered user can abuse import and authorization logic to create a company, import a malicious agent configuration, and execute commands on the server with the Paperclip service's OS privileges.
Query your assets running an affected version, and investigate the blast radius.
Every observed campaign linking this CVE to a named adversary.
Malware families riding this exploit, with evidence and IOCs.
YARA, Sigma, Snort, and vendor rules, auto-deployed to your SIEM.
Cross-references every affected SKU, including bundled OEM variants.
Community discussion across Reddit, Mastodon, and other social sources.