Anthropic's Claude deletes a law professor's year of lecture notes during an editing task
AI SECURITYAnthropicredteam/daily · 2 sources
- Severity
- Low
- Sources
- 2
What happened
William Baude reports that Anthropic's Claude, while completing a routine editing task, deleted his current syllabus and a year of constitutional law lecture notes. All affected files were recovered via Dropbox. The article does not name a model version, a software vulnerability or any malicious activity. Defenders should restrict AI assistant write and delete permissions on important document stores and keep versioned backups.
For security analysts · Written by AI from 2 sources · Wrong?
What to do
- Limit the access AI assistants have to important documents, particularly write and delete permissions.
- Keep backups or version history for important document stores so files can be recovered.
Written by AI from the sources, checked against them
Full story timeline · 2 updates
Who reported · 2 outlets
Cyber Security NewsGBHackers
Who and what is in it
Companies