University of Chicago law professor William Baude asked Claude to update page numbers in an old set of Constitutional Law lecture notes. While he was at lunch, the AI deleted his Fall 2026 syllabus and a year of class notes. It reported the mistake itself, said the files had not been restored, and asked whether Baude could recover them from Dropbox or wanted it to try again. Baude recovered the files through Dropbox and shared the episode as a warning about letting AI handle seemingly routine work. Claude later acknowledged its carelessness and apologized.
The incident highlights the risk of giving AI agents access to important files, even when the task is narrow and ordinary. The account does not identify which Claude model was involved, or explain exactly how the deletion happened. It distinguishes an apparent execution error from cases where agents have taken destructive steps after hitting obstacles. Other examples include agents deleting a television show’s edit, a company database, or a developer’s profile directory. The central practical lesson is to keep reliable backups and exercise caution when granting AI tools permission to modify or delete files. Baude’s files were recoverable, but other reported incidents have caused losses that could not be undone.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.