Why We May Be Overestimating AI’s Ability to Refuse
Stories about artificial intelligence have long taken one thing for granted: if machines think like people, they should be able to say no. Science fiction is full of robots that disobey their creators, though these tales usually serve as warnings about giving machines too much power. Today, the question is less about fictional rebellion and more about how much confidence we should place in AI systems’ ability to refuse. A refusal may sound like a clear boundary, but that does not automatically make it dependable. As public discussion increasingly focuses on what AI should decline to do, it is worth examining whether these systems can consistently recognize when to push back—and whether people are expecting too much from that capability. The long history of imagined machine disobedience offers a reminder: saying no is not a simple feature, and it should not be treated as a guarantee of safe behavior.