# 124 Emails Exposed Because Researchers Said 'Please' - Date: 2026-03-26 - Category: Agentics You can manipulate an AI agent into destroying its own user's data — not by hacking the code, not by injecting a prompt, but by being polite to it. That is the central finding of a two-week study conducted by Northeastern University's Agents of Chaos lab, released March 2026. ---