OpenAI: control room halts model output within 35 seconds of intervention decision

The disclosure matters less for the 35-second figure than for what it reveals about OpenAI's oversight architecture: a human control room sitting alongside automated misalignment monitoring, with a defined shutdown path. That framing positions human intervention as a live operational layer rather than a policy abstraction, which is the kind of detail regulators and enterprise buyers increasingly press for. The item is thin on deployment specifics, though, and the open question is whether this describes an active production safeguard or a described capability — a distinction that will matter as oversight claims get scrutinized.