The AI regulation smackdown isn’t over

At the start of the week, major AI figures appeared tentatively aligned on slowing or coordinating safety: Anthropic CEO Dario Amodei’s three-step plan (third-party evaluators in labs, domestic coordination, international agreements), with public agreement on aspects from Sam Altman, Demis Hassabis, and Elon Musk.

That unity cracked. Meta’s Mark Zuckerberg opposed limiting company autonomy; the Wall Street Journal reported Zuckerberg, Musk, and Nvidia’s Jensen Huang scuttled proposals for an industry-funded independent regulator akin to FINRA. OpenAI’s Chris Lehane told The Verge government still has an important role and that Congress should turn bipartisan momentum into mandatory national safety standards for frontier AI.

The Trump administration has called an AI safety crisis a “hoax,” with the president telling a conference crowd over speakerphone with Huang that robots will not take over, and posting that the only needed guardrails are a “STRONG AND SMART” president plus existing criminal and regulatory power. Former DHS emerging-tech policy director Nick Reese notes Trump signed EO 13960 on federal AI principles six years ago and has not repealed it—making the current anti-regulation posture a real shift.

We’re for mandatory frontier safety standards that survive CEO mood swings—and against treating “the president is enough” as a control system.

Amodei’s evaluator-and-coordination pitch looked like the adult path after hacking and eval-gaming incidents. Watching Zuckerberg, Musk, and Huang reportedly kill a FINRA-style idea shows how fast voluntary industry governance collapses when autonomy is on the table. Lehane’s line—companies start, government finishes—is the only structure that has ever worked for aviation and finance.

Useful AI scales better with shared floor rules than with vibe-based vetoes. Reese’s point lands: framing the problem as robots taking over distorts the harm already here from non-embodied systems. Exempting AI from failure-intolerant sectors’ playbook is a choice, not a law of innovation.

I’m watching whether Congress actually drafts mandatory national standards, whether Trump’s position stays “vibes over policy” as Zvi Mowshowitz put it, and whether OpenAI’s public ask survives the next week of CEO-on-CEO pressure.

The Verge published Field’s report on September 19, 2026, as part of its “AI Superintelligence Slowdown” series. Anthropic declined to comment beyond Amodei’s essay; SSI, SpaceX, and Google did not respond; Thinking Machines Lab declined; Meta pointed to Zuckerberg’s X post.