The right to stop began as an ontological-drift safeguard — a field has to retain permission to halt a form that is no longer serving, even one heavily invested in — and it becomes most concrete at the edge of loss of control. Applied to a frontier deployment it means a genuine, pre-committed ability to stop: to pause, roll back, or refuse to proceed when evidence indicates unacceptable risk, with the burden of proof placed on the developer to show a system is safe rather than on others to prove it is not.
Its hardest test is sunk cost. A lab may have spent billions of dollars, years of research, and enormous political capital, and none of that can be allowed to create an entitlement to continue against the evidence — because investment does not confer jurisdiction over irreversible global risk. The right to stop is what keeps momentum, expenditure, and prestige from silently converting into permission. It depends on two companions: the protected dissent that surfaces the reasons to use it, and the demonstrated recovery capability that makes stopping physically possible in the first place.