“Creepy” ChatGPT calls users by name—even when it shouldn’t

An unsettling glitch has surfaced among some users of OpenAI’s latest models o3 and o4-mini, where the chatbot appears to address people by name—despite them never having shared it in conversation or enabled personalisation settings.

Simon Willison, a software developer and AI commentator, posted on X (formerly Twitter) a screenshot of ChatGPT using his name in a reasoning trace. “Does anyone like the thing where o3 uses your name in its chain of thought, as opposed to finding it creepy and unnecessary?” he wrote.



Others soon chimed in with similar experiences. One user posted: “umm o3 is using my first name in reasoning traces. when did they start giving the models our names?”

While some found the quirk amusing, others labelled it “confusing” and “creepy”. Notably, in multiple cases, users said the model used their name in a reasoning step but denied knowing their name when asked directly.

A likely explanation points to ChatGPT’s memory feature, which allows it to remember user-specific details to personalise future interactions. However, many of the users reporting the issue claim to have disabled memory and customisation settings—casting doubts over how the model accessed their name.

Also Read: ChatGPT can now remember what matters to you

The glitch has reignited privacy concerns and transparency questions around AI systems. OpenAI CEO Sam Altman has previously spoken about the long-term vision of AI models that “get to know you over your life” to offer more useful and personalised experiences.