Unpatched flaw in Anthropic Claude's document skills lets attackers run code on Linux
AIR Security researchers found a vulnerability, dubbed 'Office Party,' in Claude's official DOCX, PPTX, and XLSX skills, which rely on a helper script that launches LibreOffice via a shared library loaded through Linux's LD_PRELOAD mechanism. Because the script doesn't verify the origin of an existing shared-temp-folder library file, an attacker who plants a malicious file there can get it executed with the privileges of the user running Claude the next time a document is processed. Anthropic has declined to fix the issue.
GoKawiil's interpretation of the reporting above, not reported fact.
The flaw illustrates how AI assistants that automate routine office tasks can quietly introduce new attack surfaces into enterprise systems, since the vulnerable mechanism touches a shared temp directory used by many other programs. Anthropic's decision not to patch suggests the company may view this as a lower-priority or shared-responsibility risk, leaving organizations to mitigate exposure on their own, according to researchers at AIR Security.
- AIR Security disclosed 'Office Party,' a flaw in Claude's DOCX/PPTX/XLSX skills enabling code execution on Linux.
- The bug exploits an unverified shared library loaded via LD_PRELOAD during LibreOffice-based document conversion.
- Anthropic has chosen not to fix the vulnerability, leaving affected environments exposed.
Source: darkreading.com — Elizabeth Montalbano, 2026-11-09
Published there as: “Attackers Crash Claude's 'Office Party' Flaw to Run Malicious Code on Linux”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.