Bloomberg reports that Treasury Secretary Scott Bessent helped develop a proposal for an independent AI agency reporting to the SEC — modeled on FINRA — that would screen frontier models for dangerous capabilities before release. It is a plan under review, not a policy. But the direction of travel is the story.
From Ad-Hoc Controls to Proposed Institution — Timeline
Early 2026
Trump administration temporarily imposes export controls on an advanced Anthropic model; OpenAI asked to limit a new model’s rollout. Ad-hoc posture draws complaints from Silicon Valley.
Google DeepMind CEO Demis Hassabis proposes an industry-funded, FINRA-style self-regulatory body governed by the labs themselves — a bid to set standards before Washington does.
July 17, 2026
Bloomberg reports the White House is reviewing a Bessent-developed proposal for an independent AI agency reporting to the SEC — same FINRA template, pointed the opposite direction. Voluntary 30-day pre-release submission for bioweapon, deception, and hacking screening.
Status: July 18, 2026
Under review by White House Chief of Staff Susie Wiles. Not enacted, not funded, not written into law. Submission would be voluntary, not a hard licensing gate.
Bloomberg reported on July 17 that the Trump administration is considering the creation of an independent AI regulatory agency that would report to the Securities and Exchange Commission. The proposal, which Treasury Secretary Scott Bessent helped develop, is explicitly modeled on the Financial Industry Regulatory Authority — the self-regulatory body that oversees US brokerages. It is now under review by White House Chief of Staff Susie Wiles. No legislation has been drafted, no funding allocated, and no timeline announced.
As described to Bloomberg, the body would screen frontier AI models for dangerous capabilities — specifically deception, the ability to help create bioweapons, and malicious hacking skills. Frontier labs would voluntarily submit their models for review roughly 30 days before public release. The proposal follows a period of ad-hoc government intervention: Anthropic was in negotiations with officials after the administration temporarily imposed export controls on an advanced model, and OpenAI was asked to limit the rollout of a new system. Those case-by-case actions drew complaints from Silicon Valley leaders about the lack of a formal process.
This morning we covered how the White House has been increasingly dictating which entities get access to frontier models — an ad-hoc posture with no institutional backing. The Bessent proposal, if enacted, would be the institution behind that posture: a standing body rather than a series of one-off requests. The honest bracket is wide: Bloomberg’s sourcing describes an internal proposal that may change substantially or never ship. But the direction of travel is consistent and legible.
The key insight: The significance here is not the acronym — it is the institutionalization. A standing body that vets models before release reclassifies frontier AI structurally, the way drugs or aircraft are classified: as a controlled technology with a pre-market review process. Proposing that institution is the tell, regardless of whether this specific proposal survives the review process intact.
The Structural Read
There are three structural reads on this proposal, each worth holding separately.
1. The state is proposing a permanent gatekeeping institution for the frontier.
Ad-hoc requests — export controls here, a rollout limit there — reclassify frontier AI as strategically sensitive on a case-by-case basis. A standing regulator with a pre-release review process reclassifies it structurally. This is the analytical frame from The Geopolitical Fencing of the Frontier: the frontier is being fenced, and the fence is moving from informal to formal. Proposing the institution is the tell even if the institution never ships exactly as described.
2. The real contest is self-regulation versus state regulation — and both proposals use the same template.
Three days before the Bloomberg report, Demis Hassabis proposed an industry-funded, FINRA-style self-regulatory body run by the labs. Bessent’s version borrows the identical FINRA template but points it the other way — an agency reporting to the SEC rather than a club the labs govern themselves. The analogy is precise and double-edged: FINRA is itself a self-regulatory organization that the SEC oversees, so “reports to the SEC” still leaves enormous room for the regulated to shape the rules. Who writes the capability standards — the labs or Washington — is the entire fight, and both proposals are competing bids to answer it first.
“The Permission Layer is the layer of the AI stack that governments control. When it thickens — more formal, more permanent, more institutionalized — it does not disappear as a factor in competitive dynamics. It becomes the dominant one. The question is not whether a permission layer exists for frontier AI. It does. The question is who writes its rules.”
Three Implications
IMPLICATION 1 — FOR THE LABS: COMPLIANCE COST AND TIMELINE RISK
A mandatory 30-day pre-release review window — even a voluntary one that becomes normative — inserts a new variable into every release calendar at Anthropic, OpenAI, Google DeepMind, and Meta. The labs that already invest in pre-deployment safety evaluations absorb this more easily; those that ship faster on looser internal processes face a structural adjustment. The broader risk is that voluntary norms harden into de facto requirements before any formal rule is written, as happened with financial self-regulation in earlier eras.
IMPLICATION 2 — FOR GOVERNANCE: THE CAPTURE AND CONCENTRATION CRITIQUE
The FINRA model is well-understood and carries a well-understood criticism: self-regulatory organizations, even those overseen by a government body, tend to be shaped by the largest incumbents in the regulated industry. A handful of frontier labs writing the capability standards that define what “dangerous” means is a different kind of governance than an independent scientific body doing the same. Critics on both sides of the aisle have raised this — those worried about industry capture of safety standards and those worried that incumbent labs will use safety gatekeeping to raise barriers to entry for smaller competitors. Both critiques have historical grounding and neither is dismissed by the FINRA framing.
IMPLICATION 3 — FOR GEOPOLITICS: THE OPEN-WEIGHT ASYMMETRY SHARPENS
Every formal gate placed on the US closed frontier makes the open-weight alternative more attractive to developers, enterprises, and governments outside the US who do not want to operate under an American regulatory regime. China’s open-weight models — already competitive on several benchmarks — do not pass through a FINRA-style body before they ship. If the US watchdog is perceived as a tool of export control as much as of safety assurance, the international response may accelerate a bifurcation that the proposal was partly designed to prevent. This is the same dynamic that played out with semiconductor controls, and it is not resolved by making the safety body more rigorous.
Business Engineer Framework
The Permission Layer
The Permission Layer framework maps how government controls over model access, export, and pre-release review reshape competitive dynamics across the entire AI stack. This proposal — whether it ships as described or not — is the Permission Layer becoming structural rather than ad-hoc. Understanding which layer of the stack a regulatory action touches, and who controls the writing of its rules, is the analytical frame that separates noise from signal in AI governance. The full Map of AI traces how this layer interacts with every layer above and below it.
Bloomberg’s July 17 report describes a proposal, not a policy — voluntary, under review, potentially subject to significant revision or abandonment. Hold it accordingly. But the structural signal does not depend on this specific proposal surviving intact: the ad-hoc phase of US frontier AI governance — case-by-case export controls, one-off rollout limits, informal access negotiations — has generated enough friction that both industry and government are now reaching for institutional forms. Hassabis wants the labs to build that institution. Bessent wants the SEC’s orbit to anchor it. Both bids use the same FINRA template. Whoever writes the capability standards that define “dangerous” at the frontier will have answered the most consequential governance question of the decade — and the race to write them is now formally underway.
Gennaro is the creator of FourWeekMBA, which reached about four million business people, comprising C-level executives, investors, analysts, product managers, and aspiring digital entrepreneurs in 2022 alone | He is also Director of Sales for a high-tech scaleup in the AI Industry | In 2012, Gennaro earned an International MBA with emphasis on Corporate Finance and Business Strategy.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
