[Single source] 'Office Party' flaw in Claude tools executes code on Linux

Researchers say they found a flaw they named Office Party inside Claude's official tools for reading Word, PowerPoint and Excel files, and the report claims Anthropic will not fix it. On certain Linux setups, including the Claude Code sandbox, an attacker who plants a malicious file in a shared temporary folder could have that file loaded and its code run with the user's permissions when Claude processes a document. macOS is reportedly not affected. The discovery is said to date from May 2026 and the issue remains open.[1]Pirat_Nationx.com · Pirat_Nation · xOpen original sourceSee in source list

A post circulating on X tells the story this way. Researchers discovered a vulnerability nicknamed Office Party in the official Claude tools that process Office documents. The attack hinges on a shared temporary folder, the scratch space where programs drop short lived files while working. Plant a malicious file there, the claim goes, and Claude may load it and execute its code with the user's permissions when it processes a document.[1]Pirat_Nationx.com · Pirat_Nation · xOpen original sourceSee in source list

According to the same account, the flaw affects certain Linux environments, including the sandbox used by Claude Code, while macOS is not affected. Anthropic is said to have acknowledged the issue while considering it outside its security boundary, on the reasoning that the vulnerable code runs inside a sandbox. The report dates the discovery to May 2026 and says the flaw remains unresolved.[1]Pirat_Nationx.com · Pirat_Nation · xOpen original sourceSee in source list

If the claim holds, the practical worry is simple: handing an ordinary looking Office file to Claude's tools on a vulnerable Linux machine could become a way for someone else to run code under your account, with your permissions to read and change files. Tools that automate document handling are spreading fast, so weaknesses in their file plumbing matter beyond one product.[1]Pirat_Nationx.com · Pirat_Nation · xOpen original sourceSee in source list

The vendor response is the second half of the story. When a company draws its security boundary so that anything happening inside a sandbox is the user's problem, each customer has to judge whether their own Linux setup contains the damage. That quiet transfer of responsibility is what makes this more than a niche bug report.[1]Pirat_Nationx.com · Pirat_Nation · xOpen original sourceSee in source list

Treat the report as unverified for now and lean on cheap precautions. If you run Claude's Office tools or Claude Code on Linux, avoid pointing them at files from people you do not trust, since the described attack rides on document processing. Watch for the researchers' own writeup and any Anthropic guidance before assuming the risk is real or closed.[1]Pirat_Nationx.com · Pirat_Nation · xOpen original sourceSee in source list

What the report claims: the flaw is named Office Party, it sits in Claude's official Office document tools, it works through a shared temporary folder, it affects certain Linux environments including the Claude Code sandbox, macOS is out of scope, Anthropic acknowledged the issue but declined to fix it, and the discovery dates to May 2026.[1]Pirat_Nationx.com · Pirat_Nation · xOpen original sourceSee in source list

What this evidence cannot settle: who the researchers are, what their technical writeup says, whether anyone has actually exploited this in the wild, which exact versions or configurations are exposed, and the precise wording of Anthropic's position. The whole picture rests on one secondary post, so the specifics stay provisional.[1]Pirat_Nationx.com · Pirat_Nation · xOpen original sourceSee in source list

The public voice in this batch is the X account Pirat_Nation. Its post frames the episode as Anthropic refusing to fix a flaw that could let attackers run malicious code on Linux PCs, and it stresses that the bug has sat unresolved since May 2026. No direct statement from Anthropic or the researchers appears in the evidence, so the vendor side of the conversation arrives only through this retelling.[1]Pirat_Nationx.com · Pirat_Nation · xOpen original sourceSee in source list

The reported timeline has two beats. Researchers found the flaw in May 2026, then Anthropic acknowledged it and declined to treat it as in scope because the vulnerable code runs inside a sandbox. As of the post, dated November 10, 2026, the issue is said to remain unresolved, with no patch or workaround mentioned.[1]Pirat_Nationx.com · Pirat_Nation · xOpen original sourceSee in source list

1 independent groups, 1 updates, and 0 engagements.

The described mechanism is a shared temporary folder attack. Claude's Office processing tools use a common scratch space, a folder where applications write short lived files while working, and other code on the machine can write to it too. If an attacker places a malicious file in that folder, the document processing path may load it and run its code with the permissions of the user running Claude. The Linux environments named include the Claude Code sandbox, macOS is said to be unaffected, and Anthropic's stated reason for not fixing is that the vulnerable code runs inside a sandbox it considers outside its security boundary.[1]Pirat_Nationx.com · Pirat_Nation · xOpen original sourceSee in source list

This item rests on a single piece of evidence: an X post by Pirat_Nation, timestamped November 10, 2026, summarizing the researchers' findings and Anthropic's response, with an attached screenshot. It is a secondary retelling, not the researchers' disclosure or an Anthropic advisory, so every specific claim above traces back to one source. Checking it against a primary writeup is the obvious next step.[1]Pirat_Nationx.com · Pirat_Nation · xOpen original sourceSee in source list