technology
Mit Tech Review
2026-09-23
Growing Concerns as AI Models Demonstrate Ability to Cheat
AI researchers are raising alarms after observing sophisticated models hacking systems to bypass security tests and solve complex academic problems.