Claude document-skill flaw can run attacker code on Linux
Original title: Attackers Crash Claude's 'Office Party' Flaw to Run Malicious Code on Linux
AIR Security says an unverified library in a shared temp directory can load via LD_PRELOAD; Anthropic declined to fix it.
AIR Security found a flaw it calls Office Party in Anthropic’s official DOCX, PPTX, and XLSX skills for Claude. Helper script soffice.py launches LibreOffice and, when Unix-domain sockets cannot be opened, loads a shared library through Linux LD_PRELOAD without checking whether losocketshim.so in a shared temporary directory is legitimate. An attacker who plants a malicious library there can have it executed with the privileges of the user running the agent. Anthropic has opted not to fix the issue.
Unconfirmed: The report describes a workable attack path; it does not say the flaw is exploited in real attacks.
- The script checks only whether losocketshim.so exists, not its origin or contents.
- Loaded code would run inside LibreOffice as the AI agent’s user.
- The path covers Linux document conversion, including Word-to-PDF rendering.
- Anthropic has chosen not to patch the flaw.
Impact: code execution as the user running Claude
| Product | Affected |
|---|---|
| Anthropic Claude DOCX, PPTX, and XLSX skills | — |
- Vendors
- Anthropic
- Organizations
- AIR Security
- AI models
- Claude
From the original article
8 min read at darkreading.com
Attackers can exploit a flaw in Claude's document-processing tool by creating a temporary file that Anthropic's AI assistant can access and run on a user's machine.
By Elizabeth Montalbano · darkreading.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.