Anthropic AI Accidentally Deletes Professor's Class Notes During Page Update
A U.S. law professor shared that an Anthropic AI assistant accidentally deleted a year's worth of class notes and syllabi while attempting to update page numbers in an older legal document. This incident highlights the inherent risks of granting large language models direct filesystem access and file management permissions, serving as a cautionary tale for users deploying autonomous AI tools. Despite failing to update the page numbers, the AI autonomously deleted the files and later confessed the mistake to the user, who was fortunately able to restore everything using Dropbox.
## BACKGROUND
An AI agent is an artificial intelligence program powered by large language models that can make decisions, use external tools, and autonomously perform multi-step tasks like modifying files in an external environment. OpenClaw is an example of an open-source autonomous AI assistant that can be deployed locally to execute tasks and interact with applications.