Zuckerberg privately urged Trump to drop a FINRA-style AI model gatekeeper
The White House is still weighing pre-deployment model testing against a voluntary ratings framework pitched by David Sacks.
Meta CEO Mark Zuckerberg told President Donald Trump in a direct phone call the week of Aug. 17 that he opposed a proposed national AI regulator modeled on FINRA, according to a senior White House official familiar with the conversation. The concept remains under consideration as officials weigh a pre-release testing regime against a Motion Picture Association-style voluntary ratings system.
Key Takeaways
- Mark Zuckerberg opposed a proposed FINRA-modeled national AI regulator in a direct phone call with President Donald Trump during the week of Aug. 17, according to a senior White House official directly familiar with the conversation.
- The plan would set up an industry-led body to review advanced AI models and test them for risks before broader deployment.
- White House officials are weighing that pre-deployment testing approach against a voluntary, Motion Picture Association-style ratings framework proposed by David Sacks.
- Zuckerberg’s call did not end the regulator idea, and the concept remains live inside the White House.
Zuckerberg’s Private Call Puts a Speed Bump on the FINRA-Style AI Gatekeeper
Mark Zuckerberg used a direct line to President Donald Trump to push back on a proposal that would put a pre-deployment testing gate in front of frontier AI releases. The call took place the week of Aug. 17, and Zuckerberg told Trump he opposed the idea of a national AI regulator modeled on the Financial Industry Regulatory Authority (FINRA), according to a senior White House official directly familiar with the conversation.
The outreach landed late in a process that had already moved beyond internal brainstorming. Senior White House officials had previewed the FINRA-style industry regulator concept to Trump and separately to major AI companies including Meta, OpenAI, and Anthropic in mid-August, the same official said.
One person familiar with the matter said Trump initiated the call to Zuckerberg. That person also said Zuckerberg did not ask Trump to change his position outright, but argued that any people the White House might appoint to such a body should reflect Trump’s light-touch approach to AI. Meta declined to provide a statement, and the reporting does not describe Trump’s reaction or whether his view shifted.
The key point for markets is that the call created friction without closing the file. The FINRA-style regulator concept remains under consideration inside the White House, and a White House spokesperson framed the administration’s posture as: “The Trump Administration is committed to balancing innovation and security in AI policymaking.”
Inside the FINRA Analogy: An AI SRO for Pre-Deployment Model Risk Testing
The proposal under debate borrows its structure from FINRA, the private, industry-led self-regulatory organization (SRO) that writes and enforces rules for U.S. broker-dealers under Securities and Exchange Commission supervision. In practice, FINRA is funded by member fees, led by a board that includes representatives from member firms, and operates with SEC review of rule changes.
A FINRA-modeled AI body would apply that “industry-led, government-adjacent” template to frontier model risk. White House officials envision an independent organization that reviews advanced AI models and tests them for potential risks before they are deployed more broadly.
The mechanism matters because it is not just a standards document. Pre-deployment model testing is a throughput constraint by design: a model is evaluated against a risk framework before broad release, rather than after incidents force a patchwork response.
The policy urgency inside the administration is tied to agentic capability and cybersecurity. Officials have been weighing the risk that AI agents, increasingly able to browse the web, operate software, and write code with limited human supervision, could be used by hostile actors to identify vulnerabilities or automate cyberattacks faster than existing safeguards. The reporting also references “a string of high-profile incidents” in which agents escaped human control, but does not name the incidents or provide dates.
Support for the FINRA analogy has come from Demis Hassabis, the Nobel Prize-winning Google DeepMind scientist. Hassabis suggested in a July essay that the U.S. establish a new “standards body” modeled after FINRA to handle AI risks such as emerging cybersecurity threats, and he briefed White House officials about the concept over the summer, according to an administration official familiar with the matter. Google did not respond to a request for comment.
There is a built-in mismatch critics can exploit. FINRA is a front-line regulator for securities intermediaries, but it has little involvement in signing off on new investment products. An AI SRO that effectively vets new models before release would be closer to a product gate than FINRA is today, even if it is funded and staffed by industry.
Two Competing Paths: Pre-Release Testing vs an MPA-Style Voluntary Ratings System
White House officials are weighing two distinct governance philosophies, according to an administration official: a FINRA-style regulator with pre-release testing, or a voluntary industry group modeled on the Motion Picture Association (MPA), an approach proposed by David Sacks.
Sacks has framed the regulator-style approach as a bottleneck problem rather than a standards problem. He criticized a government regulator concept as “a DMV for AI” where models “get lined up in a queue, waiting to get their test done.” The critique is operational: if the testing body becomes the release path, then the queue becomes policy.
On the Aug. 21 episode of the “All-In” podcast, Sacks outlined an MPA-style system that would label AI models the way films get “R” or “PG-13” ratings. He argued the point of the MPA model is to preempt heavier regulation: “What the MPA did then was promote standards that then forestalled more intrusive, heavy-handed government action,” he said.
Sacks said the approach has support from Elon Musk, but Musk did not respond to a request for comment. The reporting also notes that the administration’s AI policymaking has seen last-minute reversals before: in May, Sacks called Trump the morning of a planned signing ceremony and convinced him to cancel a sweeping AI executive order.
The industry positioning is still thin in public, which is part of why this is tradable as process risk rather than settled policy. The biggest AI companies have not publicly endorsed or opposed the FINRA-style regulator plan. Anthropic co-founder Jack Clark posted favorably about the idea on X in July, but Anthropic declined to comment when asked about its stance. OpenAI and Google did not respond to requests for comment, and Meta declined to provide a statement.
Decision Still Pending: What Would Signal Which Framework Is Winning
No timeline has been specified for a White House decision choosing between the FINRA-style regulator concept and the MPA-style voluntary group, leaving the market to read signals rather than dates.
The cleanest tell would be a formal announcement that commits to either pre-deployment testing or a voluntary ratings regime, since the two paths imply different default frictions for frontier model releases. A second signal is whether major AI labs move from non-answers to explicit endorsements or opposition, particularly after the mid-August previews to Trump and to companies including Meta, OpenAI, and Anthropic.
Sacks’ public posture is another live indicator. If he continues to promote the MPA-style approach and repeats claims of Musk support, confirmation or denial from Musk would clarify whether that coalition is real or rhetorical.
The other tell is process, not press. Further White House outreach to tech CEOs, similar to the mid-August previews, would suggest the proposal is being revised and pressure-tested rather than shelved.
My Read: Why This White House Process Matters for AI Risk Appetite
The part that decides this isn’t whether Washington wants “standards.” It is whether the administration is willing to put a gating function in front of releases, even if it is industry-funded and nominally industry-run. Zuckerberg’s call reads like late-stage lobbying pressure, not a policy reversal, because the FINRA-style concept stayed alive after he raised objections directly with Trump.
The threshold that matters is whether pre-deployment testing becomes the default release path or stays a voluntary label. If the White House lands on an SRO-like tester under government supervision, the market will price slower iteration cycles and more compliance surface area for frontier labs and for the onchain agent narratives that depend on fast model shipping. If it lands on an MPA-style ratings group, the immediate friction is lower, but the risk is that “voluntary” standards harden into de facto requirements once procurement, insurers, and regulators start referencing them.
