The Missing Procedure
Imagine a research project that has disclosed its methods and passed the inspections required by an agreement. One evaluation team believes it can proceed safely. Another identifies a plausible route to catastrophic harm. Both see the evidence, and both think the other has made a consequential mistake.
What happens next?
Transparency makes the activity visible. Verification establishes compliance with existing commitments. Deterrence changes the cost of defection. A decision procedure still has to establish how a disputed activity fits the rules, what evidence is sufficient, and which responses are authorized. That is the procedural question we take from 1a3orn's critique.
A complete procedure specifies who can initiate a case, who receives evidence, how parties respond, and who hears an appeal. It also specifies whether activity continues during review, when a temporary restriction is justified, and when that restriction expires. Every one of those choices distributes risk and authority.
A decision is only half the procedure. Once a legitimate decision exists, something has to make it matter, and three kinds of power can do that. Jurisdictional power binds actors that accepted the institution's authority. Access power belongs to whoever controls something others need: compute providers, chip suppliers, insurers, marketplaces, investors, and customers can each condition access on meeting a standard. Sovereign power belongs to states, through licensing, export controls, procurement, and liability law.
The institution need not hold the second or third kind. Its distinctive role may be narrower: turning evidence into a legitimate determination that independently governed actors can rely on when they exercise powers they already hold. That arrangement divides power by design. It also creates a new danger, because a determination that every cloud provider, insurer, and customer acts on at once can be as coercive as a penalty. The questions on outsiders, cartel power, and equivalence take up that danger.
The decisive test is a major participant losing. The inquiry will put participants on both sides of matched disputes: seeking protection in one, facing restraint in the other. A design has to explain why its protections remain worth accepting after the outcome becomes inconvenient.