Trump Signs 'Super Intelligence' Order as Six AI Labs Sign White House Safety Accord

President Trump signed an executive order Tuesday directing all federal agencies to replace the term "artificial intelligence" with "super intelligence" in official communications — while CEOs from Google, Anthropic, Meta, OpenAI, xAI, and Nvidia signed a voluntary White House accord committing to a four-layer safety oversight system for frontier AI development. The rebranding is rhetorical; the safety agreement is real, but voluntary.

Background

The Trump administration has positioned itself as pro-AI since day one, rolling back Biden-era executive orders on AI safety in January 2025. This accord marks a pivot: rather than deregulating entirely, the White House is now advocating for a specific industry-led safety architecture — one that bypasses Congress and the regulatory process entirely. The six signatories represent the dominant players in frontier model development in the United States.

What the accord requires

The White House safety agreement commits each signing company to implement four layers of oversight for any frontier model training:

1. Internal controls — written policies governing development, deployment, and incident response
2. Internal review team — a dedicated safety function within the company, separate from product
3. Independent external auditor — a third party assessing whether controls are working
4. Independent board committee — board-level governance for AI safety decisions

Trump described the agreement as "morally binding" at the signing. It carries no enforcement mechanism, no implementation deadline, and no requirement to publish audit results.

The terminology shift

The executive order applies to all federal agency communications — procurement documents, regulatory filings, public statements. "Super intelligence" is not a technical term in machine learning research; it conventionally refers to a hypothetical system that surpasses human cognitive ability across every domain. Applying it to today's large language models is a deliberate departure from the field's own vocabulary.

The practical effect: federal agencies will need to update templates, style guides, and procurement language. Vendors bidding on government AI contracts will face the new terminology in solicitations.

What changes for the industry

For the six signatories, the gap between current practice and the accord's requirements varies. Anthropic already publishes a Responsible Scaling Policy with internal safety checkpoints. Google DeepMind maintains a dedicated safety research team. OpenAI has a safety committee at the board level following its 2023 governance crisis.

What's new is the public, White House-witnessed commitment — and the implicit signal that companies outside the six lack equivalent oversight. No other company was invited to sign.

The enforcement gap

Voluntary tech-industry accords have a mixed record. The Biden administration's 2023 voluntary AI commitments covered similar ground — internal red-teaming, third-party audits, safety information sharing — and were later folded into executive orders but never enacted into law. The Trump accord follows the same structure: high-profile alignment with no statutory backbone.

That doesn't make it meaningless. Public commitments create reputational stakes. A signatory that skips an external audit, or refuses to disclose findings after a safety incident, faces a credibility problem that a non-signatory does not. The accord is a floor, not a ceiling — and it gives the White House a reference point if any of these companies later pushes back against more formal regulation.

What to watch

The accord's first real test: whether any of the six companies disclose third-party audit results, and on what schedule. Without a reporting requirement, "morally binding" describes intent, not accountability. Watch also for smaller AI labs to face pressure — from investors, customers, or Congress — to adopt the same four-layer structure even without signing.