Britain’s AI safety debate has reached the awkward question of whether the watchdog is guarding the house or visiting it.

The Guardian reports that company executives at a parliamentary hearing were split over mandatory government testing of new models. Despite reported U.S. pressure against sharing models with Britain’s AI Security Institute, its account of the AI minister’s response says: “Narayan confirmed this week that they had managed to retain cooperation.” The unresolved question is stronger testing requirements, not whether all present cooperation has vanished.

In our imaginary kennel, the watchdog receives a visitor’s badge. It is encouraged to investigate independently, provided it remains a pleasure to have around.

The dog asks whether the gate opens because it has authority or because the owner currently likes dogs. The handler calls this an unnecessarily adversarial distinction. Look: the gate is open.

Our fictional dog then reports an unwelcome smell. The host praises its excellent nose, offers a commemorative biscuit and asks whether the warning could be delivered with a more grateful wag.

The question is whether future testing can be required. The host proposes a badge upgrade instead: VERY CONCERNED VISITOR.