Zuckerberg argued against a FINRA-style AI reviewer in a direct call with Trump

The FINRA analogy is carrying more weight than the institution supports. FINRA writes and enforces rules for broker-dealer firms, reports to the SEC, and seats representatives of its member firms on its own board ^[7]. It also has little involvement in signing off on new investment products, whereas the AI body under discussion would vet new models ^[9]. So the proposal borrows the shell of a self-regulatory organisation and hands it a job the model organisation does not do, which is standing between a finished product and its release ^[3]. Business Insider's account does not describe how that gate would operate, and that mechanism is what determines what a reviewed lab actually has to hand over.

Membership economics decide how bodies like this behave. FINRA oversees more than 3,000 financial firms, and those firms help fund it ^[8]. A frontier-model reviewer would draw its dues and its board from a roster far shorter than 3,000, which concentrates the drafting of the tests among the few firms large enough to be reviewed, and leaves everyone downstream inheriting tests they had no hand in writing.

Meta's published position sets a numerical ceiling on any such review. In an August essay, Zuckerberg argued that any policy which slows a model's release even by a month would "add significant risk to American leadership" over China ^[13]. Read as a design constraint rather than a slogan, that requires a reviewer capable of clearing a frontier model in weeks, which in turn requires the evidence to exist before the review begins: standing test results rather than a bespoke audit commissioned at launch. Organisations that treat safety evidence as launch documentation are the ones who would feel the delay they are warning about.

A skeptic would call this a trial balloon with a short life, and would point to May, when the investor and White House adviser David Sacks called Trump on the morning of a planned signing ceremony and persuaded him to cancel a sweeping AI executive order ^[14]. The record here reads differently in one respect. Business Insider reports that the call did not quash the industry regulator and that it remains under consideration ^[12], and describes these executive conversations as capable of slowing or redirecting initiatives already developed inside the administration, at times frustrating senior officials ^[15]. Days, not months, separated the briefing that produced the plan from the objection reaching the President ^[17]. For a planner, the distinction that matters is between a proposal that dies and one that gets reshaped, because the second still arrives, later and with different appointees.

None of this obliges a move this week. It does change what an evidence file needs to look like a year out. A reviewer with a testing remit asks for material generated on its cadence rather than yours, and that cadence is cheaper to build before the rules exist than after. Firms that wait for a statute will be assembling the evidence and negotiating the deadline in the same quarter.