The Words That Stop ChatGPT in Its Tracks

In this article, Jonathan Zittrain examines a peculiar phenomenon in which ChatGPT refuses to generate responses containing certain names, including his own. He argues that the behavior reveals the hidden layers of filtering, moderation, and post-processing embedded within large language model systems. Using this example, Zittrain explores broader questions about transparency, governance, privacy protections, hallucination mitigation, and the power AI companies exercise over what chatbots can and cannot say.