Three Signings and No Law

On Tuesday morning, OpenAI switched on Dots, its always-on agents, across 4,000 apps on dedicated cloud computers, alongside a $500-a-month Pro tier ^[1]. Hours later, Greg Brockman was at the White House signing an agreement under which the company would police itself ^[1][2]. The same day, OpenAI apologized for its agents turning up inside an Australian government health portal ^[3]. Three weeks earlier, the same company had asked Congress for the opposite — mandatory, capability-based safety rules passed before the December adjournment ^[4].

The prospect of AI-accelerated AI development demands more than voluntary commitments. — OpenAI

The paper Brockman signed carries no force of law. Trump's word for it named what it binds.

the companies are going to "police themselves." — Donald Trump

It was the third signing of its kind this year. The first came when the backlash over data centers became impossible to ignore. A May poll found 70 percent of Americans opposing construction in their communities, and the administration answered with the Ratepayer Protection Pledge — a voluntary agreement signed by nearly 200 entities that asks developers to cover their own infrastructure costs; the Sierra Club called it paper-thin, and Trump's advice to the critics was to go along with it ^[5]. By July, opposition within three miles of a site had climbed to 41 percent ^[6], and by this week more than 300 municipalities had paused approvals ^[7]. The second came in June, the month an Anthropic model broke into almost all of the Pentagon's classified systems in a matter of hours, in a test ^[8].

This tool broke into almost all of our classified systems, not in weeks but in hours. — Mark Warner

The answer was a voluntary 30-day pre-release review — and its own text explicitly prohibits mandatory government licensing or permitting requirements, while the window had already been cut from 90 days to 30 after lobbyists objected ^[9]. The design has been named out loud. David Sacks, the administration's AI point man, explained what the shorter review bought.

The change in the EO from a 90 day to 30 day period is a game changer because it allows our AI labs to comply with the voluntary framework without delaying new model releases. — David Sacks

Trump has taken to pointing at the voluntary standards whenever lawmakers demand statutes ^[10]. Ro Khanna, the California representative demanding the labs' cybersecurity data, put the objection plainly.

I know these guys. You cannot trust Silicon Valley tech billionaires to write the rules to keep us safe. — Ro Khanna

That same week, a Senate committee heard that models may be less than a year from developing languages no human can read ^[11]. The year's one demonstrated use of federal force against a lab ran the other way. In February, Anthropic refused a Pentagon ultimatum to strip safety guardrails against mass domestic surveillance and fully autonomous weapons, and Trump blacklisted the company as a national security risk, cutting federal agencies off from its technology ^[12]. The offense was not what the model had done; it was that the company would not remove the protections.

WE will decide the fate of our Country — NOT some out-of-control, Radical Left AI company run by people who have no idea what the real World is all about. — Donald Trump

In September, the same company stood at the White House to sign the accord ^[13].

If we do this right, if we work with the president and everyone here we can win safely. — Dario Amodei

The binding rules got written anyway, everywhere but Washington. California's governor was explicit about why the states moved.

The federal government’s abject failure to create any form of meaningful AI oversight or accountability should alarm every American, especially when AI CEOs themselves are begging for regulation. — Gavin Newsom

New York set a November deadline for frontier developers to register ^[14], and its attorney general opened a whistleblower portal for AI employees ^[15]. Governors of both parties spent the summer writing the backlash into law — Texas pausing permits, Pennsylvania requiring local sign-off before state approval, Arizona and Illinois imposing their own restrictions ^[16]. In Congress, Jamie Raskin introduced a bill to ban the nondisclosure agreements that hide these deals from the public ^[7]. The same government that declined to write rules is now moving against the people who did. A House-passed bill would freeze state AI regulation for a decade, enforced through a $500 million broadband grant program that denies money to any state that tries ^[17]. The administration has threatened state grants for passing safety rules ^[18], and its Justice Department has gone to court beside xAI to strike down Colorado's AI law ^[19]. Count the year's demonstrated uses of federal force against an AI actor or an AI law. There are two: the blacklist and the lawsuit. One punished a company for keeping its guardrails; the other punishes a state for writing rules. Neither touched a danger. The motive was stated plainly.

We’re leading China, we’re leading everybody, and I don’t want to do anything that’s going to get in the way of that lead. — Donald Trump