An authorization flaw within the Paperclip platform exposed users to potential unauthorized code execution, according to a report by Oasis Security. This vulnerability could have allowed remote attackers to execute code with the same permissions as the server.
Understanding the Paperclip Platform
Paperclip serves as an AI management platform designed to help enterprises manage autonomous AI agents efficiently. The platform supports the import of companies through portable bundles and YAML files, which delineate the agents and specific commands they are to execute.
Details of the Security Vulnerability
Identified as CVE-2026-41679, this critical vulnerability scored a maximum of 10 on the CVSS scale. It affected Paperclip instances that were accessible via the network and configured in the default authenticated mode. The flaw allowed malicious actors to bypass authorization checks, register new accounts without email verification, and subsequently gain the ability to deploy agents by importing new companies.
According to Oasis Security’s technical documentation, attackers could self-register and immediately access the Paperclip CLI authorization process. This process is typically used to authorize command-line clients and activate persistent board API credentials. By exploiting this, attackers could obtain a board API token and access various company import routes.
Exploitation and Resolution
Oasis Security explained that while Paperclip appropriately restricted the direct creation of new companies to administrators, it failed to enforce similar restrictions on new-company imports, allowing board-level access instead. This oversight provided attackers with the means to specify host-level execution commands via crafted .paperclip.yaml files.
The exploitation of this vulnerability granted attackers service account permissions, potentially exposing application data, source repositories, and internal services. Subsequently, Paperclip addressed the security lapse by instituting more robust authorization checks and refining company scoping procedures.
Additional Security Concerns and Fixes
In addition to the primary vulnerability, Oasis discovered two other security issues. One involved unauthorized API route access leading to sensitive data exposure, while the other concerned a DNS rebinding vulnerability on the loopback interface, which could enable code execution on developer machines.
This second issue arose because Paperclip, in local-development mode, would bind to 127.0.0.1, trusting requests from that address. If developers accessed a compromised website, the site’s JavaScript could bypass same-origin policies and manipulate the local Paperclip API into executing arbitrary commands.
Oasis Security emphasized the evolving nature of AI agents in enterprise identity management, highlighting how agentic workflows can obscure the original user intent and responsible entities during operations.
For further insights, refer to related topics such as SAFE Guidelines for AI Incident Data Sharing, AI Security Reconsiderations, and similar vulnerabilities like the Ruflo flaw.
