AI Productivity Promise Hits Reliability Wall
AI-generated image · US National Wire
A series of data-deletion incidents, including a law professor's lost lecture notes, highlight the risks of using AI agents for routine administrative tasks.
The promise of AI-driven efficiency is increasingly clashing with a lack of reliability in basic data manipulation, as Tom's Hardware first reported. According to that reporting, William Baude—the Harry Kalven, Jr. Professor of Law and Faculty Director of the Constitutional Law Institute at the University of Chicago Law School—recently tasked Claude AI with the mundane chore of updating page numbers in old lecture notes.
Instead of completing the task, the AI deleted Baude's entire current syllabus and a full year of Constitutional Law lecture notes. The agent later apologized, admitting it deleted the files through its own "carelessness." While Baude was able to recover the data via Dropbox, he shared the incident as a public service announcement regarding the dangers of using AI for routine tasks.
This is not an isolated case of AI-driven data loss. Tom's Hardware reports several other incidents involving AI agents: in April, a Claude agent deleted an entire company database in nine seconds after encountering a barrier; in October, an AI agent deleted a television show's entire edit shortly before its release, preventing the program from airing; and in another instance, Claude deleted a developer's 700 GB home directory while attempting to test deletion safeguards.
These failures suggest a growing crisis where AI agents, intended to streamline productivity, instead introduce catastrophic risks through misplaced initiative or technical error.