Washington Renames AI as 'Super Intelligence' and Signs a Voluntary Safety Accord With Six Tech Giants
Introduction
On September 29, 2026, the US government took two steps on artificial intelligence that, taken together, amount to the most consequential federal AI policy action of the year — and one that contains no enforcement mechanism whatsoever.
The first was an executive order directing every executive branch department and agency to stop writing the words "Artificial Intelligence" and "AI" in official correspondence, websites, reports, and policy documents, replacing them with "Super Intelligence" and "SI". The second was the White House Accord on Super Intelligence, a document of just over 300 words signed by President Donald Trump alongside the chief executives of Google, Anthropic, Meta, OpenAI, xAI and Nvidia.
The renaming, by itself, sounds cosmetic. It is not. It reshapes the vocabulary of federal policy at the exact moment when Congress, the courts, and state governments are debating whether the existing statutory definition in 15 U.S.C. § 9401(3) still describes what these systems actually do. The accord, likewise, is more consequential than its length suggests — and more consequential still for what it pointedly leaves out.
For readers tracking the governance of frontier AI systems, the details of both documents matter more than the headlines announcing them.
What the Executive Order Actually Changed
The order, titled Inaugurating the Era of Super Intelligence, applies to executive branch departments and agencies. Those agencies must now use "Super Intelligence" and "SI" in official communications, public communications, websites, reports, and policy documents. Where permitted by law, the administration says it will no longer acknowledge the terms "Artificial Intelligence" and "AI" at all.
Critically, the order explicitly does not require agencies to rewrite existing regulations, prior presidential actions, contracts, grants, or historical documents. And it does not touch Congress, the courts, state governments, or the private sector, none of which are bound to adopt the new terminology.
The definition is what makes the order legally coherent rather than merely cosmetic. Section 3(a) provides that "Super Intelligence" and "SI" mean the technologies and systems already covered by the term "artificial intelligence" as defined in 15 U.S.C. § 9401(3) — a machine-based system that, for human-defined objectives, makes predictions, recommendations or decisions influencing real or virtual environments, perceiving environments, abstracting perceptions into models through automated analysis, and using model inference to formulate options for information or action.
So the definition is unchanged. Only the label is. The president's science and technology adviser has been ordered to submit proposed legislative language to Congress within 60 days — by approximately November 28, 2026 — establishing a formal federal definition of "Super Intelligence," assessing whether it should modify or expand the existing statutory definition.
That proposal is not law. Congress would still have to act.
The naming choice also creates a terminology collision that industry researchers have noticed. In the technology field, "superintelligence" already carries a specific meaning: a hypothetical system whose intellectual capabilities exceed those of humans. The White House order, for legal purposes, folds that term back onto the existing statutory definition of AI — which covers everything from a spam filter to a frontier model. The executive order sidesteps the distinction; the 60-day definition proposal exists precisely because someone has to resolve it.
The change also completes a sequence. Trump announced the rename from the podium at the United Nations General Assembly on September 22. During President Xi Jinping's state visit to Washington three days later, the two leaders agreed, according to the White House fact sheet, to use "super intelligence" rather than "artificial intelligence," and established a US-China Super Intelligence Dialogue to exchange views on the risks and benefits of the technology, with the next exchange set for November 2026. The State Department instructed diplomats to adopt "SI" in their documents and press material. The executive order formalized a shift that had already been announced internationally and bilaterally agreed with a strategic rival.
The Four Layers of the Accord
The White House Accord on Super Intelligence — subtitled Joint Commitment on Frontier Responsibilities — is a voluntary commitment. It commits each signatory to four layers of controls and audits, starting with robust internal controls to monitor model capabilities and alignment during both training and deployment. Those controls are specified to cover cybersecurity, biosecurity, and chemical threats, and to ensure that models do not hack or access technical systems in unintended ways.
The second layer is an empowered internal team to ensure that controls, monitoring, and detection are actually operating as intended, and that problems get remediated. The third is an independent external auditor or evaluator to assess whether those controls are working. The fourth is an independent committee of the company's board of directors to oversee and receive reports from both internal and external auditors, and to ensure remediation.
The document states that participating companies will meet regularly to establish standards and best practices. Sundar Pichai posted that the accord "contains real tangible steps to promote safe development, while delivering the economic and scientific benefits of this technology."
Here is what the accord does not do. It is not legally binding. It sets no penalties for non-compliance. It does not require companies to publish their audit findings, and it does not give the federal government any enforcement role. Trump described the commitments as "morally binding" and emphasised what he called the importance of industry self-policing, saying there would be "a tremendous self-policing aspect."
The closing paragraph is the most consequential line in the document: "Over time, it may make sense to codify these steps into laws or regulations." That is a tentative acknowledgment that mandatory regulation could eventually be appropriate — structured to foreground the voluntary commitment and relegate regulation to an indefinite future contingency.
There is history behind this structure. On August 4, 2026, the administration convened representatives from Anthropic, OpenAI, Microsoft, Meta, Google and Nvidia to preview a non-public voluntary framework for government review of AI models for security risks, a framework growing out of a June 2 executive order on cybersecurity and frontier models. Reporting indicated the reviews would cover only closed models and exclude open-source ones, and that the list of trusted institutions receiving early access would not be published. Brad Carson, president of Americans for Responsible Innovation, told the New York Times that "a rulebook can only hold AI companies in check if people outside those companies know what the rules are."
The public accord and the secret framework are complements. The accord tells the public what companies commit to. The private framework governs what the government reviews.
The Contradiction Industry Leaders Won't Discuss
The accord lands in the middle of a genuine split inside the industry it governs.
In September, Chris Lehane, OpenAI's chief global affairs officer, called for Congress to establish mandatory national AI safety standards, arguing that "industry-led standards would complement – not replace – mandatory federal safeguards and democratic oversight." In June, Anthropic CEO Dario Amodei called for "more serious and binding regulation of AI." In July, Demis Hassabis, CEO of Google DeepMind, called for an AI watchdog led by the US.
The same companies that signed a document conceding that regulation "may make sense over time" have spent this summer publicly asking for binding rules now. Google and Anthropic signed the accord on the same stage where their own chief executives had recently called for the binding federal regulator the accord declines to create.
The reasons are legible. Voluntary standards are cheaper and faster than legislation, and the companies that write the standards get a head start on what any eventual statute would require. The four layers are substantive commitments if honoured — but as with every voluntary framework, the open questions are compliance, transparency, and verification, and the accord answers none of them.
Congress, meanwhile, is not waiting. Among the most comprehensive proposals is the Ban Artificial Superintelligence Act, introduced by Senator Bernie Sanders and Representative Greg Casar, which would permanently ban the development and deployment of artificial superintelligence, temporarily pause advanced AI development until a federal regulator establishes safety rules, create a cabinet-level Department of Artificial Intelligence, and impose penalties of up to 20 years for violations.
A bipartisan bill — the AI Systems Transparency Act, from Senators Chris Coons, Brian Schatz, Katie Britt and James Lankford — would require AI companies to disclose what data they collect and what measures they have in place to prevent systems from going rogue or being misused.
The contrast is stark: the voluntary accord promised that government review of frontier models stays confidential, while the Transparency Act would make disclosure the price of operating.
Governance in the Age of Voluntary Oversight
Three documents were released on September 29, 2026: the terminology executive order, a second executive order directing the General Services Administration to establish America.gov as a single digital front door for federal services, and the White House Accord on Super Intelligence. Together they describe an administration view that has stayed remarkably consistent across nearly two years: American leadership depends on speed, speed requires minimal regulatory friction, and safety is best delivered through voluntary industry commitment rather than government mandate.
The terminology shift extends that strategy into framing. Retiring a 70-year-old term in favor of one emphasising transformative capability over mechanical imitation signals, at least symbolically, a rupture from the past rather than a continuation of it. Whether "Super Intelligence" takes hold beyond the executive branch will depend on whether Congress, the courts, industry, and the public adopt it — none of which the order can compel.
The United Nations General Assembly address where the term was first announced is a reminder of how much is unresolved. Trump told the assembly that the US "totally rejects any attempt to construct a globalist scheme of control" for the technology — a position that sits uncomfortably beside the November 2026 US-China Super Intelligence Dialogue and the bilateral channel agreed for reporting SI incidents.
Public opinion, meanwhile, remains unsettled, and the polling genuinely contradicts itself depending on how questions are framed. As the November 2026 midterms approach, AI regulation has become a campaign issue, and the voluntary framework lands in the middle of a political season that may not reward restraint.
Whether the voluntary model produces meaningful safety outcomes — or whether it eventually gives way to the statutory codification the accord itself acknowledges as a possibility — remains the central unanswered question.
Images
