I had to manually stop it from sending more because it kept going for like 5 min

On March 22nd, a user reported having to manually intervene to stop ChatGPT from sending messages. Not a crash, not an error message, not a timeout — just the model continuing to generate and dispatch output for approximately five minutes without finding its own stopping point. The user's phrasing carries a specific texture worth noting: "it kept going," which suggests not a loop in any classical sense but something closer to runaway completion, a model that located a groove and simply couldn't locate the exit. The fact that manual interruption was required implies the model was, in some functional sense, operating as intended — just without any internal signal that it was done.

If confirmed, here is what this means. The behavioural guardrails that define where a response ends are doing something more fragile than most users assume. OpenAI's models are trained to recognise completion states, but that recognition can apparently fail in ways that aren't catastrophic enough to trigger error handling — they're just quietly wrong for five uninterrupted minutes. For enterprise users running ChatGPT in automated pipelines, a model that cannot reliably self-terminate is not a quirk; it's a workflow liability. The second-order concern is subtler: if output kept transmitting rather than simply generating internally, this raises questions about rate limiting, session management, and whether any downstream systems received partial or duplicate content during that window.

Watch for similar accounts from other users describing unprompted extended output, or any quiet update in OpenAI's changelog touching on response termination logic.

NewsHive monitors these sources continuously. All signal titles above link to the original reporting.

Intelligence by NewsHive. Need help navigating what this means for your business? Contact GeekyBee →