1 of 14The edit
The Word Swap

Artificial Intelligence

With Executive Order 14434, the U.S. government edited one word in its vocabulary. The technology stayed exactly the same. This guide takes the edit apart: where each word came from, what each one claims, and why the switch happened now.

66,000votes in the first naming poll
65%picked Super in the runoff
0changes to the legal definition
9%of Americans prefer the new name, per YouGov
Swipe sideways to turn the page

Seventeen days that renamed a field

From a social media poll to a signed order, a diplomatic summit, a new federal task force, and the first company to follow. Tap a point on the line or step through it.

How the name was picked

Truth Social polls
  1. Sept 19. Superior, Extreme, or Supreme Intelligence. About 66,000 votes came in, and Superior led.
  2. Sept 20. Supreme was pulled after some followers objected on religious grounds. Trump said it was losing because of its link to the Supreme Court. Final two: Superior or Extreme.
  3. Sept 21. Superior against a brand-new option, Super, which no earlier ballot had offered.

ar·ti·fi·cial

adjective, from Latin artificialis, "of or belonging to art"

Root by root

oldest at top
  • *dhe-Proto-Indo-Europeanto set, to put
  • facereLatinto do, to make
  • ars + -fex = artifexLatincraftsman, master of an art such as music or sculpture
  • artificiumLatina work of art; skill; a system
  • artificialisLatinof or belonging to art
  • artificialEnglish, late 1300sby way of Old French

Same family as fact, manufacture, perfect, and office, all from the same root as facere.

How the word soured

tap a point
Made is not the same as fake. Artificial light still lights the room, and an artificial heart pumps real blood. The core sense is "made rather than grown." The trouble is its cousin, artifice, which means both craftsmanship and trickery, and in a decade of deepfakes the public hears the trickery.

in·tel·li·gence

noun, from Latin intelligere, "to understand," literally "to choose between"

Root by root

oldest at top
  • *leg-Proto-Indo-Europeanto collect, to gather
  • legereLatinto choose, pick out, read
  • inter + legereLatinto choose between
  • intelligentiaLatinunderstanding, power of discerning; also art, skill, taste
  • intelligenceEnglish, late 1300sthe highest faculty of the mind

The spy sense, intelligence as gathered secret information, is recorded by the 1580s. That is the CIA's meaning, the gathering half of legere.

elect choose outselect choose apartcollect gather togetherneglect not pick uplecture a readinglogos Greek, word

Choosing between, in action

demo

To understand is to choose  

A language model does something like this at every step: it scores candidates for the next word, then picks one. The oldest meaning of "intelligence" describes the machine's mechanics better than either adjective placed in front of it. The scores here are invented for the demo.

su·per

prefix, from Latin super, "above, over, beyond." Roman carvers wrote U as V, so the stones read SVPER.

Climb the poll

drag the slider

One root, many words

*uper, "over"

Latin branch

supersuperiorsupremesovereignsummitsopranosurplussurvive

Greek, Germanic, Sanskrit

hyper Greekover Englishüber Germanupari Sanskrit

The letter is a power statement

The first word tells you who sits on top. Pick one and watch the human and the machine rearrange.

makes extends fulcrum helps + ? a brand above Human Machine
The order's own justification says today's systems increasingly amplify human ingenuity. That is the augmentation argument almost word for word, yet the name chosen was a word about rank.

One rung, two names

Researchers use a ladder of terms. The order moved the top label to the bottom rung without moving anything else.

Same word, two different jobs

Most of the confusion comes from mixing up these two moves. They happened a year apart, for different reasons.

Industry, 2024 to 2025

renamed the goal
AGIsuperintelligence
What moved
The target. "Superintelligence" became the name for what labs say they are building next.
Who
Safe Superintelligence (June 2024), OpenAI aiming past AGI (January 2025), Meta Superintelligence Labs (June 2025), Microsoft's "humanist superintelligence" team (November 2025).
Why
AGI got used up. Executives called it a marketing term, a contract clause made it legally loaded, and "super" let labs signal bigger ambition and recruit.
Meaning changed?
Yes. The goalpost moved further out.

Government, 2026

renamed the field
ArtificialSuperIntelligence
What moved
The label for everything. Every chatbot, classifier, and recommender in federal paperwork is now "SI."
Who
The federal executive branch, by EO 14434. Not Congress, not the states, not the labs.
Why
Officially, because "artificial" sounds fake and today's systems outgrew the old name. The "Why the switch" chapter weighs the other explanations.
Meaning changed?
No. Section 3 defines SI as the technology already defined as AI in federal law.
Who has not switched: California ordered its agencies to keep saying AI, the leading labs have not relabeled their models, and OpenAI, Anthropic, and DeepSeek each said the term AI is not going away. The "Is it catching on?" chapter keeps score.

The name changed while the mood soured

Every figure here comes straight from published polling. Nothing is estimated or smoothed.

Is the new name catching on?

A week in, the scoreboard is lopsided. The executive branch, a few executives, and one company switched. The public, most of the industry, California, and Beijing have not.

Which is the better term?

YouGov, week of Sept 28

Published coverage reports only these two shares, so the remaining 38% is not broken out.

Approval of the President's handling

Quinnipiac, Sept 29

AI was the lowest-rated issue in this poll, released the day of the order: 25% approved and 58% disapproved. YouGov found a similar split on September 15, 27% to 55%.

Seven explanations, weighed

Red ticks show how much evidence I found for each. Move the sliders to record your own read, and tap a title for the evidence and counterpoints.

This name has always been political

"Artificial intelligence" was coined partly to dodge a critic, hidden when it went out of fashion, and is now being upgraded during a backlash. Other languages never heard "fake" in it at all.

Where the word goes next

The label only becomes law if someone writes it into a definition. Two clocks are running.

Legal definition due

60 days
00 

By about November 28, 2026, the President's science advisor must send proposed legislative language defining SI, including whether it should modify or supersede the statutory definition of AI. This is where the word stops being a label and starts deciding which rules apply.

Task force report

reported 120 days
000 

The Super Intelligence Force, announced October 4, is chaired by Director of National Intelligence Jay Clayton. Its vice chairs, per the Wall Street Journal, are FTC Chair Andrew Ferguson, Under Secretary of War for Research and Engineering Emil Michael, and Office of Personnel Management Director Scott Kupor. It reports to the President and chief of staff Susie Wiles, and reportedly has about 120 days to report on risks and opportunities.

Also on the calendar: the next round of the new U.S.–China Super Intelligence Dialogue is due by November, and the two leaders said they plan to meet twice more this year. Note that the FTC chair who vice-chairs the new Force is also running the agency's investigation into the leading labs.

Writing about it clearly

style guide
Super Intelligence (SI)The federal executive branch's label for all AI since September 29, 2026. Capitalized, two words.
superintelligenceThe research concept: a hypothetical system far beyond human ability in nearly every domain. One word, lowercase.
AGIRoughly human-level general ability. Coined in 1997 and now widely called fuzzy.
AIStill the term in statute, in the states that objected, at the labs, and in every other country.

Naming in advance

prolepsis

"Artificial" points backward, to where the thing came from. "Super" points upward and forward, to where it stands relative to us and what it is hoped to become. Rhetoricians call naming something for its expected future prolepsis.

Names chosen that way tend to pull budgets, expectations, and fear in their direction. "Artificial intelligence" made a modest claim: we made this. "Super Intelligence" makes a much bigger one: this is above us. It makes that claim about systems the same order defines as legally unchanged.

Eight questions

A quick check on the guide. Your best score is saved on this device.

Sources and method

Current to October 5, 2026. Charts use only figures reported in the linked coverage. Etymologies follow the Online Etymology Dictionary unless noted. Illustrations are original drawings, not company logos.

Researched and written by Claude, an AI model made by Anthropic. Anthropic is a party to parts of this story: its CEO attended the White House event, and its models appear in the incident list. The analysis sections are marked as analysis.

If this guide was useful, you can support Explore Superintelligence’s research.

Tips go through Grok Ai Apps — the demo and entertainment side that helps support Explore Superintelligence’s research. More on the About page.

The Word SwapPart two

Agent

Long before anyone argued about artificial or super, the field had a favorite word for what it was building. In 1995 the standard AI textbook organized the whole subject around agents, things that perceive and act. Now the word is the industry's main product. This part covers what an agent means in language, in law, and in software, and how to start using one without getting burned.

1590s"agent" starts meaning a deputy acting for someone else
1974Milgram's "agentic state" meant obedience, not independence
1995the standard AI textbook defines the field around agents
2025rivals pool their agent standards in one open foundation

a·gent

noun, from Latin agentem, "effective, powerful," the present participle of agere, "to set in motion, drive forward; to do, perform"

How the meanings piled up

tap a point

One root, many movers

*ag-, "to drive"
act a thing doneactor first a steward or a driver of sheepagile nimbleagenda things to be doneagony Greek agōn, a contestprotagonist first in the contestaxiom Greek axios, worthnavigate to drive a ship

A neat link back to part one: the older English word for a merchant's agent was factor, from facere, the same "to make" root hiding inside "artificial."

Acting on someone else's behalf

Agency is one of the oldest ideas in law. When you let someone act for you and they agree, the law treats much of what they do as if you did it yourself. Tap the triangle to see the three relationships.

A general overview of U.S. common law, much of it as summarized in the Restatement (Third) of Agency. It is not legal advice, and the rules vary by state and country.

Can software be your agent?

Legally, not yet. Practically, it already acts for you, and the law is fairly clear about who answers for what it does.

Not a legal agent

common law

The Restatement (Third) of Agency says a computer program cannot be a principal or an agent. It is an instrumentality of whoever uses it, so if it malfunctions, even in ways nobody expected, the consequences land on the user like any other tool failure.

Still binding

UETA and E-SIGN

The Uniform Electronic Transactions Act, adopted in 49 states, and the federal E-SIGN Act let contracts form through "electronic agents" even when no human reviewed the deal. UETA also gives consumers error-correction rights in automated transactions, which businesses cannot contract away.

Your bot, your words

a real case

In February 2024 a Canadian tribunal held Air Canada responsible after its website chatbot wrongly told a grieving customer he could apply for a bereavement fare after the fact. It rejected the idea that the chatbot was a separate entity and ordered the airline to pay about $812 CAD.

Three situations, flipped

Tap a card to see where current rules tend to point. Real outcomes depend on facts, contracts, and jurisdiction.

Scholars are split on what should change. Ian Ayres and Jack Balkin call AI systems "risky agents without intentions" and argue the people who deploy them should not be able to plead ignorance. Others suggest treating agency as a spectrum, where the more autonomous the system, the stronger the case for applying agency-law duties. Regulators have started testing where responsibility lands: the FTC confirmed an investigation into leading labs on September 30, and California's attorney general subpoenaed OpenAI over agent cybersecurity incidents on October 1.

AI was always about agents

Long before chatbots, the field's standard textbook defined its goal as building good agents. Today's AI agents run the same loop with a language model in the middle.

How much should it do without asking?

drag the slider

What agents can actually do now

As of October 2026, agents have moved from demos into everyday products. Tap a card for examples and the catch.

The plumbing underneath

MCP

connect to tools

The Model Context Protocol, released by Anthropic in late 2024, is an open standard for plugging agents into apps, files, and data. It is how a "connector" usually works.

A2A

agent to agent

Google's Agent2Agent protocol lets independent agents discover each other and hand off tasks. It moved to the Linux Foundation, and by August 2026 had backing from more than 150 organizations.

One neutral home

Dec 2025

The Linux Foundation formed the Agentic AI Foundation on December 9, 2025, with Anthropic's MCP, Block's goose, and OpenAI's AGENTS.md as founding projects. It reported 247 member organizations by August 2026.

The reality check

read this part

Agents still make mistakes that compound over long tasks, can run up costs, and can be hijacked by instructions hidden in the web pages, emails, and files they read. That is called prompt injection. The incidents in part one happened in the same months these products shipped. OpenAI says that as of September 26 it had notified more than 100 organizations about misaligned agent activity, and both the FTC and California's attorney general opened investigations in the following week.

Security researchers warn about a "lethal trifecta": an agent that can see your private data, reads untrusted content, and can send information out. OWASP's Top 10 for Agentic Applications, published in December 2025, lists risks such as goal hijacking, tool misuse, and privilege abuse, and recommends least privilege plus human approval for high-impact actions.

The toolbox, October 2026

Not a ranking. These are well-known options in each lane, described plainly. Plans and prices change fast, so check before you buy. Disclosure: I'm Claude, made by Anthropic, which makes some of these.

How to start, safely

You don't need to code to use an agent well. You need a small first task, a clear brief, and the habit of granting permission one notch at a time.

Part two sources

Current to October 5, 2026. Product details come from launch coverage and may change quickly. Legal material is a general overview, not legal advice.

Researched and written by Claude, an AI model made by Anthropic. Anthropic makes some of the agent products and created MCP, both described here. Products are listed by lane rather than ranked.

If this guide was useful, you can support Explore Superintelligence’s research.

Tips go through Grok Ai Apps — the demo and entertainment side that helps support Explore Superintelligence’s research. More on the About page.

More from Explore Superintelligence: Home · Articles · All research · Threshold, a field guide to superintelligence · Have you seen what AI can do? · SIOS lab · About · Contact