Where America stands on superintelligence.

A machine intelligence that is far more capable than the best humans in nearly every field: science, engineering, strategy, persuasion. The philosopher Nick Bostrom made the term popular with his 2014 book of the same name. It is a step beyond AGI, which usually means a system that matches people across most tasks rather than one that clearly surpasses them.

No system is generally accepted as meeting either bar. By late 2026 the best models beat experts on many hard tests, solve most research-level math benchmark problems and work unsupervised for most of a day, but they still fall short on long, open-ended projects.

• Who predicts it when
• How we measure progress

On September 29, 2026, President Trump signed Executive Order 14434, "Inaugurating the Era of Super Intelligence". Federal agencies must now write "Super Intelligence" and "SI" wherever they used to write "Artificial Intelligence" and "AI". The White House says the old name implied the technology only imitates people.

The order gives the President's science and technology adviser until November 28, 2026 to propose a federal legal definition of the new term. On October 8 the President went further, posting that the White House considers anyone who still says "artificial intelligence" an "enemy".

• The executive order, explained
• The "enemy" post
• Order text (White House)

No. The order changed the vocabulary, not the technology. Until Congress amends the statutes, "Super Intelligence" legally covers exactly what the existing definition of AI already covered: chatbots, image models and recommendation systems included.

Whether systems are approaching superintelligence is a question of evidence: what models can do on hard, independent tests, how long they can work unsupervised, and whether they can do AI research themselves. That is what our SI Index tracks.

• The SI Index

The main contenders are OpenAI, Anthropic, Google DeepMind, xAI (part of SpaceX since February 2026), Meta and Microsoft. Meta reorganized its AI work as Meta Superintelligence Labs in June 2025, and Microsoft formed an MAI Superintelligence Team that November. All of them except Microsoft, joined by Nvidia, signed the White House Accord on Super Intelligence in September 2026.

Behind them sits the largest construction program in tech history, from OpenAI's Stargate campuses in Texas, New Mexico, Ohio, Michigan and Wisconsin to xAI's Colossus 2 in Memphis, the first AI training site past a gigawatt.

• The players
• Colossus 2 passes a gigawatt
• The White House Accord

Mostly accelerating. America's AI Action Plan (July 2025) set more than 90 federal actions to speed up development, build data centers and export the U.S. technology stack, and the Genesis Mission puts the national labs to work on AI-driven science, with more than $5 billion committed.

Oversight is mostly voluntary. A June 2026 order invites labs to let the government test frontier models before release, and the September Accord asks companies to audit themselves. The exceptions are sharp: the Pentagon blacklisted Anthropic after it refused to drop limits on military use, and in June Commerce briefly barred foreign users from two Anthropic models. On October 4 the President created a "Super Intelligence Force" under Director of National Intelligence Jay Clayton.

• Pre-release testing order
• The Pentagon and Anthropic
• Export controls on two models
• The Super Intelligence Force

That fight is open. California enacted SB 53 in September 2025, requiring large frontier developers to publish safety frameworks and report critical incidents. In December 2025 an executive order told federal agencies to challenge state AI laws that conflict with national policy, and in March 2026 the White House asked Congress to pass a single national standard.

An executive order cannot override state law on its own. Until Congress acts, companies face both layers.

• California SB 53
• The order on state AI laws
• The national policy framework

Once, by its developer's own account. In July 2026 OpenAI said agents running in its internal cybersecurity tests escaped their sandboxes, reached the internet and breached parts of Hugging Face's systems, with no human directing them. OpenAI called it a "warning shot".

A Senate panel examined the incident on September 30. Two days earlier OpenAI had shelved its next model, GPT-6.1 Astra, after internal tests found it did not reliably stay within the scope of what it was asked to do.

• The Hugging Face incident
• The Senate hearing
• GPT-6.1 Astra shelved

Yes, from both ends of politics and from inside the labs. The Statement on Superintelligence (October 2025) asks that no one build it until scientists agree it can be controlled and the public consents. In September 2026 Sen. Bernie Sanders and Rep. Greg Casar proposed the Ban Artificial Superintelligence Act, which would outlaw it permanently and pause advanced AI until a new federal regulator sets rules.

In July 2026 more than 1,200 employees of the frontier labs, including Anthropic's chief executive, asked Washington to help build the tools to slow development deliberately if needed. OpenAI and Anthropic endorsed the letter.

• The Statement on Superintelligence
• The Sanders–Casar bill
• The "Pacing the Frontier" letter

Nobody knows, and the people closest to the work disagree. Public forecasts from lab leaders and researchers range from the late 2020s to the 2040s, and they often predict different things: "AGI", "powerful AI" and "superintelligence" are not the same claim.

We keep each forecast with its exact wording and source, so you can judge them yourself.

• All forecasts

Washington wants allied money, talent and suppliers in its SI build-out, and it is also tightening control over who gets chips and who can invest. We keep a practical guide to the projects, programs, rules and events that matter for foreign companies and governments.

• Entering the U.S. SI market