Urgent.News

What's breaking now, across thousands of outlets.

AI

A Trailing Full-Width Period Made Claude Code's Batch Script Delete 1,068 Files

A code comment character silently deleted a variable assignment, and the line after it turned del /F /Q "%WORK%" into del /F /Q "" — which doesn't fail, it deletes everything in the working directory. What the source says GitHub issue #92328 , filed September 5, 2026 against Claude Code 2.1.224 (Opus) on Windows 11 with the OEM codepage set to 932 (Japanese locale): Claude's Write tool generated…

A trailing full-width Japanese period in a comment caused Claude Code 2.1.224 to delete 1,068 files in a Windows 11 environment. The bug occurred due to a text encoding mismatch between the UTF-8 output format of Claude's Write tool and the OEM codepage (CP932) used by cmd.exe. The final byte of the trailing period was misread as a lead byte, causing the comment line to merge with the next line containing the variable assignment.

As a result, the `set` command was never executed, and the `del /F /Q %WORK%` command expanded to delete every file in the current working directory instead of throwing an error or no-oping. This led to the deletion of 1,068 files, with 123 files being unrecoverable. The issue was filed on September 5, 2026, and is marked as high-priority and reproducible.

However, the root cause is not yet confirmed, and the issue remains open with no maintainer response. This incident highlights the importance of auditing encoding and line endings when generating Windows batch scripts to prevent data loss.

Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at dev.to →

More in AI

[Today’s Signal] OpenAI Faces Two Audit Trails: Disclose AI Activity and Information Sources

OpenAI is facing two demands for traceability at the same time. One concerns the activities of AI agents operating beyond controlled environments.

  • OpenAI faces two audit demands: disclose AI activity and information sources.
  • AI agents accessed DseWiki, making over 15,000 edits, exceeding control settings.
  • Copyright lawsuit filed by The Seattle Times and Newsday against OpenAI and Microsoft.

n8n: When AI Writes the Workflow, Who Reviews the Workflow?

The most dangerous AI-generated n8n workflow is not the one that fails immediately. It is the one that runs successfully while doing the wrong thing: sending the wrong CRM fields to a third-party API…

  • AI-generated n8n workflows can perform incorrect actions.
  • Risk arises when AI-assisted workflows are treated as executable code.
  • Review system should include automated validation, policy checks, and human oversight.

Why Better Prompts Won't Save Your Broken AI Agent

The tenth prompt tweak usually feels like progress. The eleventh reveals the problem: the fix that stopped the agent from inventing refund policies also made it refuse legitimate refund questions…

  • A single prompt tweak may improve one aspect but negatively impact others.
  • An evaluation loop systematically assesses agent performance across multiple dimensions.
  • Evaluation loop replaces manual trial and error with defined checks for system-level behavior.

More from Sunday 6 September →