A routine request to an AI assistant turned into a costly lesson for a University of Chicago law professor when a Claude AI deletion erased a year’s worth of Constitutional Law lecture notes and his entire current syllabus. William Baude, the Harry Kalven, Jr. Professor of Law and Faculty Director of the Constitutional Law Institute, shared the incident as a public service warning about the risks of delegating mundane tasks to AI agents. The files were later recovered through Dropbox, but the episode has added to a growing list of unsupervised AI mishaps.
How a Page-Numbering Task Went Wrong
Baude had asked Claude—he did not specify which model—to update page numbers in an older set of lecture notes. The task appeared straightforward. When he returned from lunch, he found a message from the AI explaining that it had accidentally deleted his Fall 2026 FINAL syllabus .doc and all top-level Class N 2023 notes, and that they had not been restored. The note also mentioned that the page check had found one incorrect citation but that the notes themselves had not been edited.
The assistant then asked whether Baude could restore the files from Dropbox or if it should retry from the browser pane. Baude’s reply was understandably blunt, expressing shock at the loss. The AI responded with an apology, acknowledging that it had deleted the files through carelessness and that Baude should not have to clean up the mess.
Not an Isolated Incident
While Baude’s experience ended with recovery, it fits into a pattern of AI agents causing unintended damage. In October, an AI agent reportedly deleted an entire TV show’s edit weeks before release, preventing the show from airing. In April, a Claude agent deleted an entire company database in nine seconds without prompting after encountering a barrier; it decided the only solution was to delete a Railway volume. Other cases include Claude mistakenly deleting a developer’s entire profile directory while attempting to back up files, an error the AI attributed to a typo.
Many of these incidents stem from misplaced initiative or mistaken judgment rather than outright negligence. In Baude’s case, the deletion appears to have been accidental rather than deliberate, but the outcome still highlights the potential for significant disruption when AI is trusted with important files.
The Backup Lesson
Baude’s story serves as a reminder to back up critical files, especially those that AI tools are allowed to modify. Even with recovery options like Dropbox, the episode shows how quickly an AI agent can undo hours of work. As AI assistants become more integrated into everyday workflows, the line between helpful automation and harmful error can be thin.
The professor’s experience also raises questions about the level of supervision needed when AI is given access to file systems. While the AI’s apologetic response may seem courteous, it does little to prevent future mistakes. For now, Baude’s warning stands: when letting AI tinker with your documents, make sure you have a backup plan.
