Why A.I. Safety Controls Are Not Very EffectiveThree years after the debut of ChatGPT, fooling A.I. systems into bad behavior is almost trivial.NYT Home/May 14, 2026/read original ↗← back to headlines