On Tuesday 29 September, Donald Trump signed an executive order titled Inaugurating the Era of Super Intelligence. From now on, the US government no longer says "artificial intelligence", but "Super Intelligence", or SI for short. The same day, the heads of the largest AI companies signed an agreement on safe development. It sounds like a joke, but it is not: the order is on whitehouse.gov. What actually changes is a lot smaller than the headline suggests, though.
What exactly was decided
The order is short. The main points:
- Federal departments and agencies use "Super Intelligence" and "SI" instead of "Artificial Intelligence" and "AI" in correspondence, on websites, in reports and in policy documents. The old terms are no longer acknowledged.
- Existing regulations, contracts, grants and historical documents do not have to be rewritten.
- Legally, SI means exactly the same as AI. The order points to the existing statutory definition of artificial intelligence in US law (15 U.S.C. 9401).
- Within 60 days, the president's science adviser has to propose a statutory definition of SI of its own.
The reason, according to the White House: today's technology goes far beyond what people had in mind when the term "artificial intelligence" was coined some seventy years ago. Trump put it more briefly: "artificial" sounds fake, while it is actually amazing.
The new name was picked with a poll on Truth Social. More than 50,000 people voted and 65 percent chose "super intelligence" over "superior intelligence". Both options abbreviate to SI, so the outcome was settled before the first vote came in.
The agreement with the tech companies
After a lunch at the White House, Elon Musk, Greg Brockman (OpenAI), Dario Amodei (Anthropic), Mark Zuckerberg (Meta), Sundar Pichai (Google) and Jensen Huang (Nvidia), among others, signed the Joint Commitment on Frontier Responsibilities. Trump called it "morally binding" and "almost like a constitution".
What it says, in short:
- internal controls on risks such as cyberattacks and biological and chemical threats;
- a dedicated team that checks whether those controls work;
- independent external auditors;
- oversight from the board, and fast remediation when something goes wrong;
- models must not break into systems in unintended ways.
What it does not say: which auditor, which standard, which deadline, and what happens if a company does not live up to it. Even Mike Johnson, the Republican Speaker of the House, called it voluntary commitments.
Anthropic, the maker of Claude that our hosting runs on, signed too. Dario Amodei added that the technology has real risks and that the rules to address them are still being discussed.
Joke or serious?
Both. The order is real and civil servants have to carry it out. In substance, nothing changes: SI is AI with a different label. It fits a pattern. In January 2025 the Gulf of Mexico became the Gulf of America for the US government, and in September 2025 the Department of Defense got the name Department of War added. A new name tells a story, and here that story is that America wins the race. Trump said so himself: "Whoever wins SI, whoever wins superintelligence, wins."
The serious part is in the agreement, not in the name. A voluntary agreement without a regulator or sanctions is mostly a promise. That is not nothing: a lab that promises on paper to be audited externally can be held to that. But it is not a law, and it does not replace one.
What are the consequences?
For the US: little, for now. Government websites and reports will say SI. Laws, contracts and the statutory definition stay AI. The moment to watch comes in 60 days: if a new definition is proposed and Congress adopts it, SI could legally come to mean something other than AI. That is when it gets interesting for anyone who supplies the US government.
For Europe and the Netherlands: nothing. Here the EU AI Act applies, and it simply talks about AI systems. If you work for US government customers, you will deal with two words for the same thing. One line in your glossary ("SI = AI") is enough.
For the word superintelligence: damage. In tech, superintelligence means something specific: a system that far exceeds humans in virtually every domain. That does not exist yet. Labs say they are working on it, researchers warn about its risks. If every chatbot and every spam filter is now called SI, what word is left for the real thing? A warning about superintelligence will soon sound like a warning about your autocorrect.
For expectations: damage too. We work with Claude and Codex every day. They are impressively good, and they still make mistakes an intern would not make. Someone who expects a superintelligence checks less carefully. That is the real risk. That is why the assistant on our hosting asks for permission first before it changes anything, and on shared hosting works in a jail that only sees your account. It also fits the White House agreement: no access to systems the model has no business in.
What if you carry out the order literally
Picture a diligent sysadmin who gets the task "replace AI with SI everywhere" and goes about it the way sysadmins do.
$ echo "The mail server on this domain is maintained daily." | sed 's/ai/si/g'
The msil server on this domsin is msintsined dsily.
Programmers know this as the clbuttic mistake, after a filter that replaced ass with butt and so turned classic into clbuttic.
A few more things that have been superintelligent since Tuesday:
- The SI system. The metre, the kilogram and the second have been SI units since 1960. A kilo of apples is now a superintelligent kilo.
- The musical scale. Do, re, mi, fa, sol, la, si. Anyone who sings now sings about superintelligence.
- Spanish. Sí means yes. The White House has summed up its AI policy in one word.
- America.gov. Along with the order, the White House launched a chatbot for questions to the government. So it was superintelligent from day one.
The Dutch once had a word of their own: KI, kunstmatige intelligentie. It lost to AI, without any decree. It may not have helped that in the countryside, KI already meant artificial insemination. Language does what language wants.
What we are doing
Nothing. On this site AI stays AI, and Claude stays Claude. The model does not get smarter or dumber from it, and neither does your hosting. If a new definition arrives in 60 days that matters for European customers, you will read it here.
And the assistant on your hosting? We just call it an assistant. It promises less and delivers more.
Want to see what Claude and Codex can and cannot do in practice? Read 8 bugs we only found by using it ourselves or have a look at the Invoker Assistant.
Sources: the order on whitehouse.gov, TechSpot on the agreement, Newsweek on the poll, Fox Business, IAPP on the legal side.