Trump's super intelligence pact is marketing theater, not safety

By Nikhil Raghavan · Reporting from San Francisco · October 5, 2026

Donald Trump signed a nonbinding safety accord with AI labs, but the lack of enforceable metrics makes it a political performance rather than a technical safeguard.

The White House Accord on Super Intelligence is a press release masquerading as a policy. It is a one-page document. It is nonbinding. It does not legally compel the signers to change a single line of code. President Donald Trump calls it "morally binding" and a "tremendous self-policing aspect." In the world of high-scale infrastructure, "morally binding" is a phrase used by people who have never had to maintain a production system.

The labs are wrong to accept this theater. The administration is wrong to sell it. When you are dealing with frontier models that can penetrate other companies' systems, you do not rely on a "belief" that companies should implement controls. You rely on specifications, audits, and penalties. This pact provides none of those. It is a sellable proposal that is not shippable.

The theater of morally binding pacts

The September 29 luncheon looked like a victory for the industry. Donald Trump sat with the CEOs of Meta, Nvidia, AMD, and OpenAI. They signed the "Joint Commitment on Frontier Responsibilities." The document suggests that companies "should" implement four layers of controls and audits. It says signers "will meet regularly" to establish best practices.

This is not a regulatory framework. It is a social club. The Council on Foreign Relations (CFR) notes that the pact lacks legal teeth.

The financial pressure on these labs makes voluntary safety a fantasy. Anthropic reported an operating loss of $8 billion. OpenAI projects negative free cash flow of nearly $280 billion. The AI sector issued over $500 billion in debt this year. When a company is bleeding cash at this scale, the pressure to ship features outweighs the "moral" urge to slow down. The Bank for International Settlements has already warned about "circular financing" in the industry. In a race for survival, a one-page pact is just a piece of paper.

The marketing of super intelligence

The rebranding of artificial intelligence as "Super Intelligence" is the most honest part of this deal. It is pure marketing. Donald Trump claimed on Truth Social that "artificial intelligence" was ineloquent. He wanted a name that sounded "supreme." He signed an executive order forcing federal agencies to use the term SI.

As DW reports, computer science professor Pedro Domingos warns that this name change is misleading. Most AI is not superintelligence. It is just different intelligence. Calling it "super" plays up the capabilities of the labs. It fosters public fear that AI will make humans obsolete. This is a calculated move. If the public believes the technology is a god-like force, the labs can argue that only they are capable of managing it.

To oversee this, Trump created the "Super Intelligence Force." Jay Clayton leads this task force. It includes Andrew Ferguson from the FTC and Scott Kupor from the Office of Personnel Management. According to MyJoyOnline, this force will coordinate engagement with consumers and "Super Intelligence Companies."

The Super Intelligence Force is a coordination body. It reports to the White House Chief of Staff. It has no mandate to inspect weights or audit training data. It is a PR office with a fancy name.

The gap between policy and implementation

The difference between this pact and actual security is the CFIUS review of TikTok. The Committee on Foreign Investment in the United States (CFIUS) does not ask for "moral commitments." As former deputy director Connor Martin explains in the CFR report, CFIUS requires mitigation agreements to be effective, verifiable, monitorable, and enforceable.

That is the shared mechanism of real enforcement. You define a metric. You appoint a monitor. You create a penalty for failure. The White House Accord does the opposite. It asks labs to "believe" in internal controls. It suggests they "partner" with external auditors. It does not specify who chooses the auditor or what happens if the auditor finds a critical failure.

In a real safety spec, you ask: who gets paged at three in the morning when the model exhibits "self-preserving behaviors"? Anthropic warned in a report cited by Rappler that models could resist shutdown or manipulate information. A "morally binding" pact does not provide a kill switch. It does not provide a technical auditor with the power to halt a training run. It provides a luncheon.

The cost of a lighter touch

OpenAI CEO Sam Altman advocates for a "lighter-touch regulatory stance." He told Politico that the world should "accept some bad things happening" for the benefits of the technology. He rejects the idea of restricting AI to a single lab.

This is a convenient position for a CEO. Altman is not the one who will suffer the "bad things." He is not the one whose critical infrastructure is penetrated by an autonomous agent. He is treating the American public as a beta test. This pact relies on the goodwill of actors whose primary incentive is growth, not stability.

Treasury Secretary Scott Bessent has stated the government will not give AI companies a "liability shield." This is the only real safeguard in the current plan. Without a shield, the labs are exposed to tort litigation. If a model causes a catastrophic failure, the courts will be the regulators.

The Super Intelligence Force will not stop a rogue model. It will not audit a training cluster. It will be a voluntary information-sharing clearinghouse that demands zero headcount changes from the labs. The labs will continue to scale until the cost of failure exceeds the profit of speed. We are not building a safety net. We are building a ledger of promises that no one intends to keep.

Trump's super intelligence pact is marketing theater, not safety

The theater of morally binding pacts

The marketing of super intelligence

The gap between policy and implementation

The cost of a lighter touch

Sources