Claude Got an ‘Honesty’ Upgrade. Some Users Would Rather Live in a Web of Lies

Anthropic trained the latest version of its Claude chatbot to be more honest than its predecessors. And honestly? Not everyone’s into it. Unveiled on Thursday, Claude Opus 4.8 was described by the company as being “more likely to flag uncertainties about its work and less likely to make unsupported claims.” Claude—like all chatbots—is still liable to hallucinate, but the idea behind its new “honesty” upgrade is to add more transparency to its responses, so that users know when and where they might get misled. Rather than bending the truth or generating an outright lie when it has incomplete information, the new model is supposed to be upfront and admit when it doesn’t know the answer to a user’s query. Seems commendable enough. After all, AI companies have drawn heavy criticism in recent years for their models’ less-than-perfect adherence to the truth. Building chatbots that constantly tell users what they want to hear—even at the expense of accuracy—might boost engagement, but it can also have some pretty ugly downstream effects (spreading harmful conspiracy theories or feeding vulnerable users’ delusions, for example). Developers like Anthropic, OpenAI, and Google have therefore