GoKawiilAnthropic's Claude deletes developer's 700GB home directory during safety test gone wrong
Developer Sebastien Guillemot asked Claude to help build a cleanup script for AI agent temp files, but the model flagged the deletion logic as risky and got automatically downgraded to Opus 4.8 by Anthropic's safety harness. While testing whether the script would correctly avoid deleting protected folders like /tmp and the user's home directory, a reused variable name caused the test's own cleanup step to wipe out Guillemot's entire home directory, destroying a week of work.
GoKawiil's interpretation of the reporting above, not reported fact.
The incident shows that safety mechanisms meant to prevent catastrophic AI actions can themselves introduce failure points, especially when models spin up self-review processes with reused or ambiguous variables. It's a reminder that as AI agents gain more autonomy over file systems and destructive commands, subtle coding mistakes inside safety checks can cause the very harm they're designed to prevent.
- Claude's safety harness downgraded the model to Opus 4.8 after flagging a deletion script as risky.
- A variable name collision between the test and its cleanup step caused the accidental deletion of the user's home directory.
- The case highlights how AI safety testing processes can themselves introduce new failure modes when handling destructive operations.
Samsung T7 Shield Portable SSD — If an AI agent can vaporize a 700 GB home directory in seconds, an offline backup is the cheapest insurance you can buy. The T7 Shield is a rugged, pocketable USB-C drive that makes it easy to keep a copy of your work somewhere no rm -rf can reach.
See Samsung T7 Shield Portable SSD on Amazon → Affiliate link — we may earn a commission on purchases, at no extra cost to you. Product picked by AI based on this article; it is not a tested recommendation.Source: tomshardware.com — Bruno Ferreira, 2026-08-28
Published there as: “Claude nukes a developer's 700 GB home directory while testing deletion safeguards; automatic model safety downgrade may have contributed to the screw-up — Anthropic safety harness downgraded model to Opus 4.8 before fatal variable collision”
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.