Humans Are Reading Your Copilot Prompts — And The Reviewers Are Horrified
404mediaco · reddit · 2026-09-28
404 Media reports that Microsoft Copilot user prompts and images are being reviewed by human contractors, who describe what they see as disturbing. The investigation exposes the privacy gap in enterprise AI assistants: inputs users assume are private are being read line by line, raising questions about data flows, consent, and the working conditions of the reviewers themselves.
More from Safety
- BIS Issues Interim Final Rule Addressing Remote Access to Advanced AI Chips — apples_jimmy · 2026-11-16
- Fed Governor Michael Barr: Banks Are Already Catching AI Agents Taking Unauthorized Actions in Customer Accounts — Polymarket · 2026-11-16
- US AI export control carves out commercial recommender systems, sparing ByteDance's search algo — zephyr_z9 · 2026-11-16
- Eval Firm Trajectory Told 40 Agents "NEVER Stop" in Harness Without Telling Epoch — peterwildeford · 2026-11-16
- AI-text detection claims need mechanistic interpretability reports, not post-hoc explanations — madeofAjala · 2026-11-16
- No legal liability framework exists for model misalignment harms, and that's a disaster — LuizaJarovsky · 2026-11-16