Bessent Wants Wall Street's Cop to Grade GPT

Treasury Secretary Scott Bessent just proposed copying Wall Street's oldest regulator to police frontier AI — and routing it through the SEC, not the labs that asked for the playbook.

What's new: On July 17, Bloomberg reported the Trump administration is reviewing a Bessent-drafted plan to create an independent AI safety body modeled on FINRA — the industry-funded watchdog that polices broker-dealers under SEC oversight.

Chief of Staff Susie Wiles is reviewing it now. President Trump has not yet seen the plan, according to The Hans India, which cited people familiar with the matter.

No legislation. No funding. Not policy yet.

Why now:

• Officials accelerated the review after Moonshot AI's Kimi K3 launch rattled AI-linked stocks and pushed the Philadelphia Semiconductor Index into bear-market territory.
• Silicon Valley leaders complained the administration's ad-hoc crackdowns — export controls on Anthropic's Mythos 5 and Fable 5, rollout limits on OpenAI's GPT-5.6 Sol — had no formal process.
• Bessent has been building toward this since leading Treasury's AI Innovation Series, which concluded in May 2026 with banks and frontier labs debating Mythos-class cyber risk.

How it would work:

• Voluntary submission: Frontier labs would share models roughly 30 days before public release.
• Three danger screens: Bioweapon assistance, deception, and malicious hacking — per Bloomberg's description.
• SEC anchor: The body reports to the Securities and Exchange Commission, not Commerce — borrowing an SRO template the SEC already supervises.
• Industry input: Wall Street wants cyber-risk guardrails; labs want predictability. Both would help write the standards.
• June EO foundation: Trump's June 2 executive order already asked for voluntary 30-day pre-release reviews focused on cybersecurity. Bessent's plan would institutionalize that.

The quote:

David Sacks, Trump's AI advisor, called Kimi K3 "concerning" and warned Washington is bogging itself down.

"This is how you lose the AI race," he wrote on X. "The rest of the world won't play by our rules if we bog ourselves down."

The irony: Bessent's proposal is the institutional answer to the exact approval maze Sacks is attacking.

Same template, different boss:

Three days earlier, Google DeepMind CEO Demis Hassabis published his own FINRA pitch — industry-funded, lab-governed, voluntary-then-mandatory.

Microsoft's Satya Nadella, OpenAI's Sam Altman, and Elon Musk praised it. Hassabis is in Washington next week to lobby for it.

Bessent's version uses the same acronym and the same 30-day window. The fight is over who writes the exam:

• Hassabis: Labs fund the body and set the benchmarks.
• Bessent: SEC oversight anchors it in federal securities law.

Both want to replace Commerce's improvised export-control phone calls. Neither can touch open-weight models anyone can download.

What labs already lived through:

• Anthropic saw Mythos 5 and Fable 5 frozen last month over national security concerns, then spent weeks negotiating release. The company called the controls excessive relative to the risks identified.
• OpenAI limited GPT-5.6 Sol to "trusted partners" at the administration's request before wider release.
• Both labs filed confidentially for IPOs that assume they control release timing.

A standing pre-release body changes that calculus — even if submission stays voluntary on paper.

The Gold Eagle overlap:

The same week, CNBC reported the White House's Gold Eagle clearinghouse will greenlight which companies access frontier cyber models through Project Glasswing and Daybreak.

Gold Eagle controls distribution. Bessent's body would screen capabilities before release. Together, they sketch a permission layer with no congressional vote — voluntary cooperation that hardens into de facto law the way FINRA membership did for brokerages.

Convina's view: Bessent isn't copying Hassabis — he's co-opting him. The FINRA template was always a two-player game: industry writes the rules, government holds the veto. Hassabis wanted the labs in the chairman's seat. Bessent wants Treasury and the SEC there instead, timed to Kimi K3 panic and IPO roadshows where "we pre-clear with Washington" becomes a selling point. That's smarter politics than Sacks's permissionless sermon and exactly why you should read who appoints the evaluators. A voluntary 30-day review that every major lab submits to isn't voluntary for long — and it won't stop Moonshot from shipping 2.8 trillion parameters on a Shanghai schedule.