What Is SI? Super Intelligence and the White House Accord

On September 29, 2026, President Trump, House Speaker Mike Johnson and the heads of Nvidia, Anthropic, OpenAI, Meta, Google and xAI signed the White House Accord on Super Intelligence: A Joint Commitment on Frontier SI Responsibilities. The same day, the administration declared that "AI" is now officially "SI."

This post defines the term, summarizes what each leader said in two public appearances, and states what the accord does and does not obligate. It is based on transcripts of the White House press event and a companion America.gov appearance with Elon Musk and Jensen Huang, cross-checked against press coverage of the accord.

What SI Means

SI stands for super intelligence. President Trump announced the change at the press event: "It's not AI. It's SI… We've changed the name officially today." He called "AI" a "fake word." Speaker Johnson and every executive on stage used the new term, and Musk corrected himself mid-sentence ("AI, which SI, pardon me").

Two points matter for enterprise readers.

It is a naming decision, not a technical threshold. In research usage, super intelligence means systems that exceed human performance across most cognitive domains. At the event, the term was applied to today's frontier models. Anthropic's Tom Brown described the new Opus 5.5 release as continued progress in "IQ points for each of these super intelligences." The label describes the commercial category, not a capability test.

It sets the vocabulary for regulation. The accord uses "SI" throughout. Policy documents, procurement language and future legislation may follow, so compliance and risk teams should expect to map "SI" to whatever their internal policies call frontier AI.

Jensen Huang extended the vocabulary to infrastructure: data centers, he said, "are really SI factories" because "you're producing something of economic value. You're just not storing it."

What the Accord Commits Companies To

Speaker Johnson described the accord as "voluntary." Coverage of the text describes four layers of control for each signatory:

Pichai compared it to "financial controls in a company." Zuckerberg said boards of directors would "independently review the reports that come from the auditors." Huang called the standard one "every single company ought to embrace." Signatories also acknowledged that the steps may later be codified into law.

The accord imports the audit architecture of financial reporting into frontier AI: internal control, independent review, board accountability.

Asked whether it is binding, Trump answered: "I think it's morally binding." The accord carries no statutory penalty, and critics have called self-policing inadequate. Speaker Johnson said Congress would keep "steady hands at the wheel" and continue to deliberate.

What Each Leader Said

Three tensions ran through the questions from reporters. Reporters asked whether self-regulation is sufficient, and the answer was yes from the President, and "a start" from Zuckerberg. Amodei was asked directly about self-policing and answered on risk and winning safely rather than endorsing the framing. Job displacement drew Musk's "age of abundance" response. Local data center opposition drew commitments from Trump and Johnson that companies would pay for their own power and share benefits with host communities.

The Infrastructure Case

The second video, an America.gov appearance, put numbers on the build-out. These are speaker projections, not audited figures.

• Power scale. Musk said US average consumption is about 500 GW, so each 5 GW of new generation is roughly 1%. He asserted that a 1% rise in power tracks a 1% rise in GDP.
• Orbital compute. Musk said SpaceX and Tesla aim for 200 GW of solar per year, with launches carrying "a few hundred gigawatts a year" of compute. He said space solar delivers nameplate output, against a fifth to an eighth on the ground.
• China. Musk said China has "about three times the electricity production of the United States" and named power as the long-run constraint.
• Agent safety. Huang described a containment-and-monitor pattern: agents run in an isolated sandbox (OpenShell), and a separate chip (BlueField) monitors them out of band. "You should never trust that it's contained."

What It Means for Enterprise Operators

The accord binds developers, not deployers. Enterprises consuming SI systems remain accountable to their own regulators. The EU AI Act and sector rules such as SR 11-7 continue to apply regardless of a voluntary US accord.

Vendor assurance becomes a procurement input. Expect requests for evidence of external audit and board review from signatories. Contract language on audit access, incident notification and model change control is now easier to justify.

Jurisdictional divergence widens. A US voluntary, industry-led model differs from the EU's statutory approach. Multinationals need an operating model that satisfies both, which is the core argument in Sovereign AI infrastructure planning and geopolitics.

Codification is signaled. The accord itself anticipates future law. Building auditable, documented controls now is cheaper than retrofitting them later.

The Structural Read

SI is a rebrand of a category that already existed, paired with a governance template borrowed from corporate finance. The label changes little; the template matters. It sets a reference standard for what "responsible frontier development" means, and enterprise buyers will use it as a benchmark.

What remains unresolved is enforcement. A voluntary, morally binding accord depends on the signatories' own controls and the boards that review them. For operators, the practical response is not to wait for statute but to hold their own deployments to the same four layers: controls, oversight, independent review and board accountability.