Three Demands for AI Rules, Answered With a Task Force and a Hotline

If AI is to be democratic, the most important decisions cannot be made by labs in San Francisco alone. — Sam Altman

Sam Altman, OpenAI's chief executive, said that from the Security Council floor last month, with Anthropic's Dario Amodei beside him calling AI the most important security question on earth ^[1]. A week later the two men made the ask concrete, calling for independent testing of frontier models before release — Altman through outside auditors, Amodei through the nonprofit METR ^[2]. The men running the field's two leading labs were asking Washington to put an outside check between their models and the public. Ten days after António Guterres stood before the same institution and urged the United States and China toward Cold War-style dialogue over AI, the answer arrived — not a treaty but a leader-to-leader hotline for reporting AI accidents, in the shape of the nuclear arms-control channels that came before it, announced at a summit between Trump and Xi Jinping ^[1][3]. International regulation had been waved off at that same UN session as a globalist scheme. That exchange was the middle of three demands for binding rules in under a month — from the labs, from the UN, from the states — and each drew the same answer from the same drawer. On September 9, OpenAI formally asked Congress to pass mandatory, capability-based safety requirements before adjourning ^[4]. The answer was a voluntary White House accord whose four oversight layers the companies run themselves — a covenant Trump pronounced morally binding — and on the question of slowing down, Trump had been explicit at the signing ^[5].

We are pacing our progress, which includes sometimes not training a model. — Sam Altman

The states walked up to the same door. California ordered a panel to design mandatory kill switches; New York's RAISE Act requires registration and quarterly catastrophic-risk assessments ^[6]. The first federal blow is still anticipated: the Justice Department is expected to challenge the state orders as conflicting with federal policy, using preemption — the federal power to void state laws it deems incompatible. What did arrive this week is the Super Intelligence Force: intelligence, Pentagon and Treasury officials reporting directly to the president ^[7]. He described the posture in the language of custody.

We will not in any way hinder or stifle the Growth of this incredible Industry. Rather, we will cherish it, help it, and watch over it, as it grows! — Donald Trump

Jay Clayton, who will lead the force, supplied the reason there would be no rules to slow the race.

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

Set the month's output side by side by function and it reads as a custody program. Name: an executive order puts "Super Intelligence" into every federal document and orders a formal federal definition within sixty days, and Elon Musk adopted the term for his own SpaceX unit within days ^[8]. Watch: the task force, which seats the same Pete Hegseth who in January put frontier models on classified Pentagon networks and lifted restrictions on automating nuclear decisions ^[9] — and when OpenAI found a 15,000-account campaign it attributes to China's Moonshot AI trying to extract its models' hidden reasoning, it reported the theft to government channels, not to a regulator ^[10]. Coordinate: the hotline. What the list has none of is a rule with the force of law. The task force only advises; the accord's four layers are run by the companies themselves ^[7][5]. The only binding accountability left sits with state attorneys general ^[11]. Critics, for their part, read the labs' own push for regulation as a way to raise the barrier for smaller rivals ^[2]. Return to the summit that opened the hotline, and the compass is inverted. Beijing has already written its rules into law — mandatory content labels and personal liability for executives ^[3]. Xi Jinping's demand of Trump was flat.

AI must be kept under human control. — Xi Jinping

Trump, at that same table, saw only this.

I think I’m seeing tremendous self-policing. And they understand that they have to self-police. — Donald Trump

A week earlier, from the General Assembly, he had dismissed the risk itself as a hoax that benefits China ^[1]. The country that once pushed the world toward AI rules now names the technology, defines it, watches it, and coordinates it over a hotline. The rules themselves are being written in Beijing.