CVE-2026-73032 is a remote code execution vulnerability in PapersGPT for Zotero 0.6.1 caused by unsafe use of window.eval() on unsanitized JavaScript code returned from an LLM endpoint. Attackers can exploit this by injecting malicious prompts in PDFs, intercepting API requests, or providing a malicious LLM endpoint, resulting in arbitrary code execution within Zotero's privileged context. This allows attackers to read/write files, execute processes, and access all Zotero data. The vulnerability has a CVSS 4.0 score of 9.4, indicating critical severity. No patch or official fix has been disclosed as of the publication date.

Exploitation of this vulnerability allows remote attackers to execute arbitrary JavaScript code in Zotero's chrome-privileged context, leading to full compromise of the Zotero environment. This includes the ability to read and write files on the host system, execute arbitrary processes, and access all user data managed by Zotero. The impact is critical due to the high level of access and control gained by an attacker without requiring privileges or user interaction beyond triggering the malicious code.

Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. Until an official fix is available, users should avoid using untrusted LLM endpoints, be cautious with PDFs containing prompts processed by PapersGPT, and monitor for suspicious activity. No official remediation or temporary workaround has been published by the vendor at this time.