The access argument is not a forecast. It is a bargaining position, and it costs nothing to hold. Pharma runs the identical play whenever a market proposes pricing rules: we will not launch here. It is said often and executed rarely, because a market you exit is one you cannot reenter on your own terms. Count how many firms have followed through. That number is the input.
Worth adding that a third option is already running while this argument continues. DSIT's AI Growth Lab call for evidence ran from 21 October 2025 to 7 January 2026, and the advisory version launched with legal services as the first sector, applications closing 27 September 2026. That is regulator-by-sandbox rather than regulator-by-statute. My view is that it weakens the access objection more than either side here concedes, because a sandbox already grants bespoke conditions firm by firm, only without a standing body anyone can appeal to or take to court. What it does not settle is what happens in a sector that never applies.
Sensible recommendations. Firewalling AISI from a regulator addresses the access concern without giving up on oversight. The EU AI Office faces a similar dilemma: it needs lab cooperation for evaluation while also holding enforcement powers.
The access argument is not a forecast. It is a bargaining position, and it costs nothing to hold. Pharma runs the identical play whenever a market proposes pricing rules: we will not launch here. It is said often and executed rarely, because a market you exit is one you cannot reenter on your own terms. Count how many firms have followed through. That number is the input.
Worth adding that a third option is already running while this argument continues. DSIT's AI Growth Lab call for evidence ran from 21 October 2025 to 7 January 2026, and the advisory version launched with legal services as the first sector, applications closing 27 September 2026. That is regulator-by-sandbox rather than regulator-by-statute. My view is that it weakens the access objection more than either side here concedes, because a sandbox already grants bespoke conditions firm by firm, only without a standing body anyone can appeal to or take to court. What it does not settle is what happens in a sector that never applies.
Sensible recommendations. Firewalling AISI from a regulator addresses the access concern without giving up on oversight. The EU AI Office faces a similar dilemma: it needs lab cooperation for evaluation while also holding enforcement powers.