Washington Trades AI Rules for Custody

In December, the White House was finishing a framework for overseeing the most advanced AI models. Its own papers described a transition from private company testing toward formal government supervision, safety-testing protocols included ^[1]. At a signing ceremony on October 2, the president stated the position outright.

regulation and governance inherently impede innovation — Donald Trump

The people who build the technology were saying the opposite days earlier, in the same building where the president renamed the subject. While he stood at the United Nations General Assembly dismissing global oversight, OpenAI's Sam Altman brought the builders' warning to the Security Council.

We could lose control of the future to AI. — Sam Altman

The year in between reads as a series of swaps. Each time someone offered the government a rule, the government answered with an instrument of custody instead. In July, Demis Hassabis of Google DeepMind proposed a watchdog modeled on FINRA — the Wall Street body that writes and enforces rules for its own industry, funded by its members — to run mandatory pre-release safety reviews and replace the case-by-case government interventions already in motion ^[2]. The White House's AI advisor answered with a flat refusal.

there will not be an FDA for AI. — Sriram Krishnan

The case-by-case system did not pause to mourn. Under a June vetting order, federal officials were approving access to OpenAI's newest model customer by customer ^[3].

at the request of the US government, it is launching today in limited preview instead of the open access launch we were planning on. — Sam Altman

In August, 1,367 frontier-lab researchers and executives published an open letter asking for a U.S.-led international effort to pace development — licensing, registration, enforceable agreements, coordination with China ^[4]. More than twenty nations and the UN Secretary-General made a parallel ask at the General Assembly ^[5]. None of it got an answer. What the administration assembled in those same months was the other kind of instrument: ownership stakes in thirty strategic companies across the chip and steel supply chain ^[6], and a Defense Secretary announcing a four-star command for autonomous weapons by October 2027 ^[7]. The one federal bill that would have let the government block an unsafe model release — and let a company challenge that block in court — stalled in September against the midterm calendar ^[8]. The bill offered rules with a judge; the vetting order offers clearances without one. Three weeks later, the administration appointed its AI czar at a signing ceremony, with no binding standard attached ^[7]. The clearest picture of custody came in February. The Department of War designated Anthropic a "supply-chain risk" — a label from arms procurement normally reserved for foreign adversaries — after the company refused to remove its own guardrails against mass domestic surveillance and autonomous weapons. Rivals that accepted allowing any lawful military use kept their access ^[9]. Anthropic sued over First and Fifth Amendment retaliation. The Justice Department's answer reduced the whole matter to three words.

These actions are unprecedented and unlawful. The constitution does not allow the government to wield its enormous power to punish a company for its protected speech. — Anthropic

Jay Clayton, the Director of National Intelligence now running the Super Intelligence Force, put the doctrine in a single sentence.

Not being first increases the identified, and unidentified, risks, particularly from our adversaries. Being first will better enable us to address those risks on behalf of the American people. — Jay Clayton

By that logic, custody is safety: the risk is not what the technology does but who has it first. Regulation contains; this administration means to hold. The newest instruments arrived in the past two weeks, and they are the least finished. At the General Assembly on September 22, the president announced a new name for the technology ^[5].

The United States also totally rejects any attempt to construct a globalist scheme to control for the artificial intelligence being spoken of so much now, here and after officially called super intelligence, changing the name, and that the use of the word artificial makes intelligence fake. — Donald Trump

On October 4 he made it an executive order binding on the whole branch ^[10]. The Super Intelligence Force that will carry it is co-led by the FTC chairman, an Under Secretary of War, and the federal personnel chief, reports to the president and his chief of staff ^[11], and was advertised with a recruiting line.

Only High I.Q. individuals need apply. — Donald Trump

Musk rebranded SpaceXAI to SpaceXSI within hours, changing nothing about the product ^[10]. The word games are not new: at the Energy Department, the administration restricted the words "sustainability" and "climate change" in the same period it dismantled the climate research center and repealed the endangerment finding — the vocabulary went first, then the authority ^[12]. None of this is finished. Only Musk rebranded; the market still ships "AI" — Gemini, Comet, Claude ^[10]. The Super Intelligence Force is a 120-day advisory task force, not an operating agency ^[11]. The Pentagon kept using the blacklisted Claude regardless of its own ban ^[9]. OpenAI's IPO is still priced for a commercial market. This is architecture under construction, not a completed seizure. But the direction is legible from the record. The builders spent the year asking for binding rules. Washington spent it refusing each one and building clearances, blacklists, equity stakes, a controlled vocabulary, and a force to administer them. In December the White House promised a transition toward formal government supervision ^[1]. It has delivered that transition, in an unexpected shape. On the access process now running in place of rules, one builder had already named the danger.

We don't believe this kind of government access process should become the long-term default. It keeps the best tools from users, developers, enterprises, cyber defenders and global partners who need them. — OpenAI