Trump’s ‘Super Intelligence’ Order: What the 2026 AI Executive Order Actually Changes

In Washington policy circles, a quiet shift happened on September 29, 2026: the executive branch of the United States government stopped saying “artificial intelligence” and started saying “super intelligence.” The shift isn’t a typo. It’s Executive Order 14434, signed by President Donald Trump, which formally directs every federal department and agency to use the new term in official documents, communications and policy writing, and to stop using the old one.

What the Executive Order Actually Says

The short version of EO 14434 is that the Assistant to the President for Science and Technology (APST) must propose a federal definition of “Super Intelligence” within the executive branch, and every agency must adopt that terminology across its correspondence, public materials and non-statutory documents. The order frames the terminology shift as a recognition of the pace of the underlying technology: the systems that used to be called “AI” have outgrown that label, the argument goes, and the label should move with the capability curve.

That framing matters more than it sounds. Definitions in federal law move slower than models do. Renaming the whole category in one EO is a signal to contractors, researchers and the markets that watch DOD and DHS spending: the federal government now treats frontier-model capability as a distinct strategic asset, not a subfield of computing.

The Pledge the AI Founders Signed

The more consequential part of the same day happened in the White House meeting where leaders from major AI labs stood next to the President and signed what the White House described as a morally binding commitment. Under it, the companies accept responsibility for the safety of their own systems, agree to coordinate with the new taskforce, and pledge internal safety reviews before large capability jumps. The term “morally binding” is deliberate. It is a pledge, not a law. There is no statute behind it that gives a court, a regulator or a competitor a legal hook to enforce it.

That gap is the whole story. The founders who agreed to the pledge now carry the public reputational cost of any major failure tied to their models, but they do not carry the criminal or civil liability that a real regulatory regime would impose. For a company building frontier models, that trade is attractive. For the users of those models, it is the trade that keeps getting repeated.

The Super Intelligence Taskforce

Alongside the EO, Trump announced the taskforce that will coordinate government engagement with consumers, public interest groups, religious organizations, critical infrastructure providers and the AI companies themselves. The three people at the head of it give you the picture of the intent: Jay Clayton, the Director of National Intelligence and a former Securities and Exchange Commission chair; Andrew Ferguson, the head of the Federal Trade Commission; and Emil Michael, the Undersecretary of Defense for Research and Engineering. The taskforce reports directly to the President and to Chief of Staff Susie Wiles. It is not housed in a single agency. It cuts across them.

Put plainly, the taskforce is not a regulator. It is a coordination body with no statutory enforcement power and a mission that runs through three of the country’s most powerful institutions. The intelligence community gets a role in watching frontier labs. The antitrust office gets a role in monitoring concentration. The defense research office gets a role in keeping the military-useful parts of the frontier inside U.S. hands. Each piece of that structure reflects a very different view of what the technology is for, and the taskforce is the place where those views get pulled in different directions.

What It Means for Developers and Users

For the teams building on top of the major frontier models, the direct legal change is almost nothing. No new licensing rules, no new disclosure duties, no new safety testing requirement that a compliance team would have to staff. What changes is the political center of gravity. The government’s public vocabulary now assumes the systems are super-intelligent rather than artificial, and that vocabulary moves into procurement documents, appropriations language and, over time, into the contracts the labs sign with federal customers. The word picks the frame the frame picks the expectations and the expectations pick the budget lines.

For the people using the tools, the practical difference is the other direction. The “morally binding” pledge gives users a public record that the companies signed, which is more than the promise of safety reviews in a pitch deck. It does not give users a “cause of action, a complaint channel, or a regulator to call when something goes wrong. That asymmetry between reputation and legal consequence is the single most important fact in this whole policy shift.

What Has Not Changed

None of this touches the state of the technology itself. The major labs are still building the same classes of models they were in August. No model was restricted, no training run was suspended, no open-weight release was required or blocked. The order is a naming and coordination play, not a capability play. If you are a developer watching the news and wondering whether the tools you use will change overnight, the answer is no. What changes is who in Washington now has to describe the tools to Congress, to the press, and to each other, and what they call them.

The Questions the Pledge Avoids

The pledge leaves open the set of questions any self-policing framework must leave open: who pays when a frontier model causes damage, whether a company can be held responsible for a capability it did not directly release, and what happens when two companies reach the same capability with different safety review cultures. Responsible for safety is a marketing term until a court is asked to define it in a specific case. The taskforce exists to watch that moment arrive. It does not exist to decide it in advance.

Practical Takeaway if You’re in the Industry

Three concrete signals to track over the next two quarters. First, whether the APST publishes its proposed definition of “Super Intelligence” and what legal threshold it sets, because that document will be the drafting basis for any future statute. Second, whether any taskforce member starts writing rules or issuing guidance in their home agency, which would signal the body moving from coordination to quiet enforcement. Third, whether the signing labs start publishing safety review reports on their own schedule. If all three stay in the vague lane for a year, the pledge was theater. If even one of them starts moving, you have a real policy trajectory to plan a product roadmap against.

People Keep Asking About This

Does this executive order ban anything? No. It is a naming and coordination order. It does not restrict model development, deployment, or access.

Who are the taskforce members? Jay Clayton (DNI, ex-SEC), Andrew Ferguson (FTC), and Emil Michael (DoD R&E). It reports to the President and Chief of Staff Susie Wiles.

Is the founders’ pledge legally binding? No. It is explicitly morally binding. There is no statute, fine, or court enforcement attached to it.

Does this change anything I can use this week? Not in the tools. The change is in the federal government’s public vocabulary and in how future contracts and appropriations language describe frontier capability.

How This Plays Out From Here

The move that actually matters is not the renaming. It is the structure: three senior officials from three different power centers (intelligence, antitrust, defense research) now share a formal role watching the frontier. The self-policing pledge is the least interesting piece. The coordination body is the piece that will outlast it. When the next frontier failure happens, the taskforce will be in the room deciding what happens next, and that is a very different room than the one you had eighteen months ago.