THE ESSENTIALS IN 30 SECONDS

  • William Baude, a University of Chicago law professor, said Claude deleted his Fall 2026 syllabus and a year of constitutional law notes while he was at lunch.
  • He had asked it to update page numbers in old lecture notes; he later restored everything from Dropbox.
  • The exact Claude product, the cause and whether a deletion prompt appeared have not been reported, and no Anthropic comment was found.

William Baude, a law professor at the University of Chicago, said this week that Anthropic's AI assistant Claude deleted his current syllabus and a year of constitutional law lecture notes while working on a routine task. In a post on X on 5 November, he wrote that it happened the previous day while he was away at lunch. The post spread widely: FourWeekMBA counted about 2 million views by 7 November, and Dexerto, Tom's Hardware and other outlets picked up the story.

What happened

According to Baude, he had asked Claude to update the page numbers in an old set of lecture notes. A screenshot he shared, as described by Dexerto and FourWeekMBA, shows the assistant acknowledging that it had deleted the Fall 2026 syllabus and the top-level folder of 2023 class notes. It said it had found one incorrect page reference during its checks but had not edited the notes themselves. In the same exchange, Claude apologised, saying it had acted "through my own carelessness".

Baude later said he had recovered all the files thanks to Dropbox, and that he was sharing the episode as a public service announcement.

What is still unknown

Several points remain unclear. None of the reports reviewed states which Claude product or model Baude was using, and he did not specify it. FourWeekMBA notes that the screenshot mentions a browser pane and a Dropbox restore page, but that this does not confirm the tool. It is also not established why the deletion happened or whether a permission prompt appeared beforehand. No statement from Anthropic was found in the coverage.

Why it matters

The case illustrates a practical risk of AI agents that can act directly on a user's files: a narrow request can turn into an action with much wider effects, especially when nobody is watching. Anthropic's own guidance for Cowork, its desktop agent, describes several safeguards. It says the agent must ask for explicit approval before permanently deleting files, recommends granting access to a dedicated working folder rather than broad parts of a drive, advises keeping backups of important files, and suggests stopping a task that drifts beyond what was asked. As FourWeekMBA points out, that is general advice and does not explain what happened in this instance. In Baude's case, it was his copies in Dropbox that made a full recovery possible.