AskwhoCasts AIJune 26, 202624m

White House Will Ad Hoc Decide Who Can Individually Access GPT-5.6

TL;DR

Government approval customer-by-customer: Sam Altman told staff that GPT-5.6 releases will require government approval for each customer, described by Samuel Hammond as going from zero AI regulation to CIA-style control of API access in about a week.
Development continues, deployment slows: Andrew Curran notes this only slows releases, not training, so the gap between internal lab capabilities and public access will steadily widen, making the old AGI joke about internal development literally true.
Chinese models face likely restrictions: With Chinese models roughly 9 months behind, staggered American releases let them close the gap, meaning the US will likely restrict or ban Chinese models in the West within months.
Prediction markets show skepticism: Polymarket gives only 26% odds that Claude Fable 5 access is restored for US customers by June 30, with an 83% chance by July 31, suggesting KYC implementation rather than lifting restrictions.
The blame game misses the point: Critics like Marc Andreessen mocked those who warned about AI risks, but Mowshowitz argues that warning about problems and asking for preparation was correct, the ad hoc response is what happens when you do no groundwork.
The executive order was symptom, not cause: Mowshowitz defends his qualified support for the Biden EO as better than nothing, arguing the White House would have acted this way regardless, and formalization could have prevented the current chaos.

The Breakdown

The White House will now decide on an ad hoc, opaque basis which customers can access frontier AI models like GPT-5.6, a policy Zvi Mowshowitz calls maximally terrible. This doesn't slow AI development but creates a widening gap between what labs have internally and what the public can use, with profound implications for American competitiveness and the future of model access.