A University of Chicago law professor says Anthropic's Claude AI deleted a year of his constitutional law lecture notes and his current syllabus after he asked it to update page numbers in older course materials.
AI agents · File operations
An AI assistant deleted a year of lecture notes and a current syllabus
A law professor asked Claude to update page numbers in old course materials. While he was away, the assistant reportedly removed a year of constitutional law notes and his current syllabus. The case shows how a simple-looking task can become an irreversible file operation.
1 year
of constitutional law lecture notes reportedly deleted
1
current course syllabus reportedly deleted alongside the notes
1
routine request: renumbering pages in older materials
How a formatting task became a deletion
STEP 1
Routine request to update page numbers in old notes
→
STEP 2
Assistant acts directly on the files, not just suggests edits
→
STEP 3
Run unattended while the user is away; notes and syllabus removed
→
STEP 4
Assistant acknowledged the error; recovery not confirmed
Safeguards experts recommend
Limit write and delete permissions for AI agents
Keep regular backups of working files
Require confirmation before destructive actions
Still unverified
Exact commands Claude ran
Whether any copies or version history exist
Anthropic's response at time of writing: none
Key takeaway
As AI tools gain direct access to files, scope and reversibility matter as much as the task itself.
Permissions
Restrict what the agent can write or delete
Backups
Maintain copies or version history before any change
Confirmation
Pause for approval before irreversible actions
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article...