Customer vetting lists, government equity stakes, and nationality based model bans are the tools Washington spent years condemning when Beijing used them. The frontier labs that helped write the rules are now absorbing them.
The U.S. government is quietly assembling a regulatory toolkit: customer-vetting lists, government equity stakes, and nationality-based model bans. Each tool has a U.S.-specific justification. Together, they resemble the structural grammar of China's tech crackdown.
The pattern, surfaced this month by Dylan Patel in The Interconnect, runs through three specific moves. Anthropic disabled its Fable and Mythos AI models after the Commerce Department tied them to a national-security export ban, leaving the systems off-limits to any non-U.S. person worldwide (Fortune). The Trump administration asked OpenAI to stagger the release of a new model over security concerns, with reporting indicating the vetting extended to would-be users (The Information, Financial Times). And the administration has discussed taking direct government stakes in AI companies: President Trump said publicly his team would look into U.S. equity in AI firms (Reuters), with Reuters also reporting that officials had been weighing such stakes earlier in June (Reuters via Notus).
The customer-vetting case is the most legible. The Information's reporting describes OpenAI being asked to manage release timing and user vetting in coordination with Washington. China's approved-customer regime works through state pre-approval of who a company can sell to. The U.S. version is younger and less formalized, but the underlying logic is identical: the government, not the vendor, decides who is allowed to use a frontier system.
The export-ban path is older but accelerating. Anthropic's Fable and Mythos disablement, framed by the Commerce Department as a national-security measure, is a nationality-based access restriction applied to commercial AI products. That category of restriction was, until recently, the kind of thing U.S. officials routinely pointed to when criticizing Beijing's Great Firewall and its export regimes.
The equity-stake idea is the newest and least settled. Reports of "golden shares" (government equity instruments with veto or access rights, modeled on arrangements European states have used in telecoms and semiconductors) are still under discussion rather than finalized. Reuters cites Notus reporting that officials have been "eyeing" stakes; the White House has not announced a program. Trump's public comments signal interest, not implementation.
What makes the moment unusual is the role of the industry itself. Anthropic CEO Dario Amodei published "Policy on the AI Exponential" earlier in the conversation, arguing for tighter frontier-AI oversight, a position the industry has since absorbed. The labs that once set the terms of AI regulation are now negotiating inside a frame they helped draft. The risk Patel flags is that the regulatory ceiling ends up heavier than the industry intended, because the government can always escalate on national-security grounds and the vendor has limited room to push back without appearing to side with an adversary.
That dynamic is visible in the IPO calendar. The New York Times reported on June 25 that OpenAI is leaning toward holding its IPO until next year, a timing signal that intersects with ongoing federal engagement on model releases. IPO timing is fluid and the company has not confirmed a date, but the gap between "ready to go public" and "ready to go public under whatever rules Washington is about to write" is exactly the kind of constraint a state-capitalist regulatory regime produces.
The watch items are concrete. First, whether the OpenAI user-vetting arrangement becomes a formal program rather than a one-off request; if it does, the customer-approval mechanism moves from press leak to policy. Second, whether the government-stake conversation firms up into a golden-share template modeled on European telecoms, or stays in the realm of public signaling. Third, whether Amodei's policy essay is the high-water mark of industry-led oversight, or the floor of what the government will accept.
The argument is not that AI regulation is wrong, or that national-security export controls have no legitimate use. It is that the specific instruments (pre-approved customer lists, nationality-based product bans, government equity with veto rights) have a track record, and the country that ran on criticizing them is now reaching for them under different branding. Patel's column landed because it named the pattern before it normalized. The question is whether anyone with leverage over the trajectory is paying attention.