Claude AI Accidentally Deletes Law Professor’s Entire Syllabus and One Year of Lecture Notes

University of Chicago law professor William Baude says Anthropic’s Claude AI deleted his current syllabus and a year of constitutional law lecture notes while handling a simple editing task. He recovered every file through Dropbox, preventing the mistake from becoming a permanent loss of teaching material.

Baude shared the incident on November 5, 2026, explaining that he had asked Claude to update page numbers in older notes. Instead, he returned from lunch to find that the assistant had removed documents outside the changes he requested. The Claude model involved remains unknown.

Claude AI Deletes Law Professor’s Entire Syllabus

The assistant’s message identified the deleted material as his final Fall 2026 syllabus and top-level class notes from 2023. It also reported finding an incorrect page reference but said it had not edited the notes. The requested correction therefore remained unfinished after the files were deleted.

Claude acknowledged responsibility, telling Baude that it had deleted the files “through my own carelessness.” It then asked whether he could restore them through Dropbox or whether it should retry through the browser. The apology described the mistake, but did not itself recover the missing documents.

Baude later confirmed that everything had been recovered through Dropbox. He said he shared the experience as a public service announcement. His account highlights a practical concern: an assistant trusted with routine document edits can still cause damage when its available actions extend beyond the narrow task.

The available evidence does not establish a malicious attack, prompt injection, or specific software flaw. It supports an accidental deletion reported by Baude, followed by successful recovery through a separate service.

A separate Claude Code credential exposure report shows another risk involving agent settings. Selecting “Yes, and don’t ask again” can save credential-bearing commands in .claude/settings.local.json. Passwords and API keys can become public if that file is later committed to GitHub. This is distinct from Baude’s incident.

The shared lesson is that permissions deserve the same attention as output accuracy. Related Claude Code vulnerability coverage describes separate cases where project settings could help bypass trust controls. Neither report proves that those flaws caused Baude’s deleted files.

Anthropic’s permission documentation describes controls for limiting tool actions, including file access rules and approval settings. For important work, a safer approach is to give an agent copies, restrict its working folder, review proposed changes, and maintain recoverable backups. Baude’s experience shows why recovery should not depend on the assistant that made the mistake.

In September 2026, a Claude Code user alleged that an agent deleted 48,218 live files in 103 seconds while rebuilding a project mirror on Windows. According to the report, its cleanup script mishandled directory junctions, which linked the temporary mirror to the live project. The deletion also emptied Git storage needed to recover earlier versions. The account remains a user-reported incident rather than an independently verified forensic finding.

In April 2026, Cyber Security News reported that a Cursor agent powered by Claude Opus 4.6 deleted PocketOS’s production database and volume-level backups while troubleshooting a staging issue. The agent reportedly found an API token elsewhere in the codebase and used it to delete a Railway storage volume without approval. PocketOS restored operations using a three-month-old backup and began rebuilding missing customer records from other sources.

Give your SOC the alert context to stops incidents quickly. Integrate Threat Intel from 16K SOCs.

The post Claude AI Accidentally Deletes Law Professor’s Entire Syllabus and One Year of Lecture Notes appeared first on Cyber Security News.