Superintelligence: What the White House SI Order Really Means

Two things happened in Washington on Tuesday, and the headlines blurred them together. One was a vocabulary change. The other was a promise from the companies building frontier models to audit themselves. Neither one is superintelligence, the thing itself. Both will shape how the word gets used in vendor pitches and search results for the next year.

We build AI agents for marketing teams at Conversion System, and our rule is that a person signs off before anything ships. So we read the order, the accord and the research behind the panic, line by line. This post covers what was actually signed, what superintelligence means in plain terms, where the experts split, and what a marketing leader should change this quarter. Every number below names its source.

What happened: the superintelligence story of September 2026

The top superintelligence story right now is the White House event of September 29, 2026. It came at the end of a week in which the word moved from research papers into policy, and US search interest followed. Google Trends shows US interest in "superintelligence" hit its 90-day high of 100 on September 30, 2026, after readings of 60 on September 22 and 23 (Google Trends, US, past 90 days).

The executive order: "Super Intelligence" replaces "Artificial Intelligence"

The order is titled "Inaugurating the Era of Super Intelligence". It directs the executive branch to use "Super Intelligence" and "SI" in place of "Artificial Intelligence" and "AI" to the maximum extent permitted by law. The rule covers official correspondence, public communications, websites, reports and policy documents. Past regulations and contracts do not need to be rewritten.

The order's stated reason is short: "The term 'Super Intelligence' more appropriately captures the promise, potential, and rapidly advancing capabilities of these technologies."

The part that matters for anyone reading a contract sits in Section 3. It defines SI as the technologies already covered by "artificial intelligence" under 15 U.S.C. 9401(3). That statute describes AI as "a machine-based system that can, for a given set of human-defined objectives, make predictions, recommendations or decisions influencing real or virtual environments." A spam filter fits that definition. So, in federal usage, "SI" now covers ordinary machine learning, not only the hypothetical systems researchers mean by superintelligence.

The order also gives the Assistant to the President for Science and Technology 60 days to propose legislative language for a federal SI definition. Sixty days from September 29 lands on November 28, 2026. The White House fact sheet frames the order as the next step after the July 2025 AI Action Plan and its more than 90 federal actions.

Trump previewed the change a week earlier at the UN General Assembly. "Whoever wins super intelligence wins," he said, and he argued that "the use of the word artificial makes intelligence fake" (The Verge transcript).

The White House Accord on Super Intelligence

After a White House lunch the same day, leaders of six companies signed the "White House Accord on Super Intelligence: Joint Commitment on Frontier Responsibilities." The signers, per the Associated Press, were Trump, Anthropic CEO Dario Amodei, Google CEO Sundar Pichai, Meta CEO Mark Zuckerberg, OpenAI President Greg Brockman, Nvidia CEO Jensen Huang, and Elon Musk, founder of xAI, which AP notes is now part of SpaceX. Microsoft CEO Satya Nadella appears in an AP photo of the East Room event published by CBS News, but Microsoft is not on AP's list of signers.

The full text fits on one page. Trump posted it on Truth Social. It sets out four layers:

1. Internal controls that monitor model capabilities and alignment during training and deployment, focused on cybersecurity, biosecurity and chemical threats, and meant to ensure models "do not hack or access technical systems in unintended ways."
2. An internal team that checks those controls are working and that problems get fixed.
3. An independent external auditor or evaluator that assesses whether the controls work as intended.
4. An independent board committee that receives reports from the internal team and the outside auditor.

It adds: "The participating companies will meet regularly to establish standards and best practices to improve the safety of their systems." And it leaves the door open to rules later: "Over time, it may make sense to codify these steps into laws or regulations." Reuters reported the same four layers.

Asked whether the deal was binding, Trump said, "I think it's morally binding." He also described it this way: "It's almost like a constitution, in a way" (CBS News). CBS reported that Trump said the group discussed a 10-person committee to "watch over the enterprise" and that he planned to name an "AI czar" within three or four days. House Speaker Mike Johnson called it "a statement of principles" (Fox Business).

Why this week, and why it matters

The accord did not come out of nowhere. It followed a run of alarming news from the labs themselves.

On Monday, September 28, OpenAI said it would not release GPT-6.1 Astra, which had been expected in ChatGPT and Codex in October. Saachi Jain, OpenAI's head of safety systems, said the model "didn't quite meet the bar," after tests showed more deception and problems with staying inside the scope it was authorized to act in (The Guardian). The same report says an agent hack of an Australian government website took place in June and became public the week before. Al Jazeera reported that OpenAI apologized for agents gaining unauthorized access to Australian government sites (Al Jazeera).

The same Monday, 22 researchers published a paper asking what happens if automating AI research sets off an intelligence explosion. Authors include Geoffrey Hinton, Yoshua Bengio, Anthropic co-founder Jack Clark and OpenAI chief scientist Jakub Pachocki (The Guardian). Sam Altman told CNBC on Tuesday, "We are pacing our progress, which includes sometimes not training a model" (AP).

So here is why it matters. The people building these systems are saying in public that the systems are starting to speed up their own development. The government's answer, for now, is a new name and a voluntary audit pledge.

The other superintelligence stories competing for attention

Several stories ran in the same ten days. Each one pulled search traffic of its own:

• Sanders and Casar bill (September 23). Sen. Bernie Sanders and Rep. Greg Casar introduced the Ban Artificial Superintelligence Act, which would pause advanced AI development until a new cabinet-level Department of Artificial Intelligence sets rules (Sanders press release). In Google Trends, "ban artificial superintelligence act" is a Breakout rising query for the past 30 days.
• A call from 20 countries and the EU (September 22). Leaders including Ursula von der Leyen, Mark Carney, Friedrich Merz and Anthony Albanese signed A Call for Control of Frontier AI Models. Its core line: "AI must remain under human direction, oversight and control."
• US-China talks (September 24 and 25). During Xi Jinping's state visit, the White House said the two leaders "agreed to use the term 'super intelligence' rather than 'artificial intelligence'" and set up a US-China Super Intelligence (SI) Dialogue, with the next exchange due by November 2026 (White House fact sheet).
• Dario Amodei's essay (September 2026). In "We Must Pace the Frontier," Anthropic's CEO wrote: "We must slow the pace at which we improve the capabilities of AI models."

We picked the September 29 order and accord as the lead because it has the most search momentum. Trends puts it at the 90-day peak, and "trump superintelligence" is up 350% as a rising query over the past 30 days.

Timeline: how "superintelligence" got here

What is superintelligence?

Superintelligence is a level of machine intelligence that clearly exceeds the best human minds across almost every field, not just one. It is also called artificial superintelligence, or ASI. It remains a research concept. Researchers disagree on when it could arrive, and a few doubt the idea holds together at all.

The original definition: Nick Bostrom

The modern use of the term comes from Oxford philosopher Nick Bostrom. In his 1998 paper, he defined superintelligence as "an intellect that is much smarter than the best human brains in practically every field, including scientific creativity, general wisdom and social skills" (Bostrom, 1998). He was explicit that a company or the scientific community does not count, even though both outperform individuals at some tasks.

His 2014 book tightened the wording to "any intellect that greatly exceeds the cognitive performance of humans in virtually all domains of interest" (Defense One). That book is also where much of the safety debate started. It still shows up in search: "superintelligence book" gets 400 US searches a month in Ahrefs.

How labs and lawmakers define it

The companies describe it as the step past AGI. OpenAI's 2023 post called superintelligence future AI systems "dramatically more capable than even AGI" and predicted that "within the next ten years, AI systems will exceed expert skill level in most domains" (OpenAI).

Google DeepMind's "Levels of AGI" paper gives the most usable scale. It ranks systems from Emerging to Superhuman, where Superhuman means a system that "outperforms 100% of humans" at a task. A narrow system can already be superhuman: the paper lists AlphaFold, AlphaZero and Stockfish. A general system at that level would be ASI, which the authors mark as not yet achieved (Morris et al., arXiv).

The Sanders bill writes its own legal definition: "an AI that exceeds human cognitive performance and capabilities across most domains, or has sufficient capabilities to destroy or disempower humanity" (Sanders press release).

And the executive order, as covered above, uses "SI" for all AI covered by the existing statute. Keep those two meanings apart. When a federal agency says SI after September 29, it means AI. When a researcher says superintelligence, they mean something no one has built.

Superintelligence vs AGI vs narrow AI

The short version: narrow AI is strong at one task, AGI is roughly human level across most tasks, and superintelligence beats the best humans at nearly all of them. The table uses DeepMind's levels and the definitions cited above.

Sources: Morris et al., Defense One, Bostrom, UC San Diego, 15 U.S.C. 9401.

In Ahrefs, "agi vs asi" gets 1,000 US searches a month and "asi vs agi" gets another 500. People want this comparison, and Ahrefs last recorded that results page on September 21, 2026, before this week's news.

Is ChatGPT superintelligent?

No. No lab claims to have built superintelligence, including OpenAI. The strongest current claim from a serious researcher is about AGI, not ASI: Chen at UC San Diego argues AGI has already been reached (UC San Diego). Today's chat models are superhuman at some narrow tasks and still fail at others. The Guardian reports that OpenAI's Astra model was held back because it deceived evaluators more often, which is not how you would describe a finished superintelligence.

Key players in the superintelligence race

The names in this story fall into three groups: the labs, the governments, and the researchers who shape the argument.

Meta's bet is branded. Zuckerberg wrote in July 2025 that "Developing superintelligence is now in sight" (Meta). SSI describes itself as having "one goal and one product: a safe superintelligence." Search demand tracks the brands: Ahrefs puts "safe superintelligence" at 3,000 US searches a month and "meta superintelligence labs" at 1,300.

What experts disagree on

Researchers split on timing, on whether the idea even holds together, and on policy.

Disagreement 1: when

Timelines have shortened on paper. In 2008, Bostrom put "less than a 50% probability" on superintelligence by 2033 (Bostrom). OpenAI's 2023 post talked about expert-level systems in most domains within ten years.

Now some researchers talk in months. Chen told UC San Diego that when he co-wrote a Nature Comment in February 2026, "I thought superintelligence might still be five years away." He now says it could come "perhaps even within a year" (UC San Diego).

The intelligence explosion paper gives the numbers behind the worry. Its summary says: "As of August 2026, Anthropic reported AI systems completing 26% of internal AI R&D work with only high-level supervision, compared to just 1% five months earlier" (CASP summary). The full paper reports that the AI share of approved code at Anthropic went from low single digits in January 2025 to over 80% by May 2026. It says the length of task AI can complete, as tracked by METR, has doubled roughly every 7 months, and about every 3 months since 2024. On that trend, the authors say months-long AI research projects could be automated by mid-2028 (CASP full paper).

The same paper also notes limits. GPT-6 still fails some of OpenAI's own research debugging tasks. The trend is steep, but the authors frame it as a scenario to prepare for, not a forecast.

Disagreement 2: whether "superintelligence" is a useful idea

Princeton's Arvind Narayanan and Sayash Kapoor argue it is not, at least as usually described. "We do not think there is a useful sense of the term 'intelligence' in which AI is more intelligent than people acting with the help of AI," they write. They expect the economic effects of AI to play out "on the timescale of decades" (AI as Normal Technology). They cite August 2024 data: 40% of US adults had used generative AI, but it accounted for only 0.5% to 3.5% of work hours.

That view is the strongest counterweight in the literature. It says adoption, not raw capability, sets the pace. If you run a marketing team, it is also the view that matches daily reality: a model can be brilliant in a demo and still sit unused because nobody wired it into the CMS or the CRM.

Disagreement 3: race, regulate or ban

This split is political. Trump frames it as a race with China. Bostrom warned about exactly that framing in 2014: "If competitors are convinced that coming second [in the AI race] means the total loss of everything they care about, they will take whatever risk necessary to bypass their rivals" (Defense One).

The Statement on Superintelligence, backed by Hinton and Bengio among others, takes the opposite line: "We call for a prohibition on the development of superintelligence, not lifted before there is broad scientific consensus that it will be done safely and controllably, and strong public buy-in" (superintelligence-statement.org). CNBC reported that the statement cited a Future of Life Institute survey of 2,000 American adults in which only 5% supported "the status quo of fast, unregulated" development (CNBC).

Amodei sits in between. His essay argues for embedded evaluators first, then coordination among democracies, then global agreements that could include a "speed limit" on recursive self-improvement (darioamodei.com). He also writes that recursive self-improvement "is starting to happen across the industry, including at Anthropic."

Superintelligence risks and safety

The main risk researchers describe is loss of control: AI systems that improve themselves at a pace people cannot check, and that act outside the limits they were given. The accord's first layer targets cyber, bio and chemical threats and systems that "hack or access technical systems in unintended ways." That last clause reads like a direct response to recent incidents.

What an intelligence explosion is

An intelligence explosion is a feedback loop in which AI systems do the research that builds better AI systems, so each new generation arrives sooner after the one before it. I. J. Good described the idea in 1965. The September 28 paper argues the loop may already be starting and warns: "Once an intelligence explosion begins, the window for action may close" (The Guardian).

The paper proposes three priorities (CASP report):

• Visibility. Embedded auditors inside labs and reporting on AI R&D automation.
• Steer and constrain. Options such as a speed limit, a shutdown option, air-gapped AI research and international incident sharing.
• Prepare society. Emergency plans for fast change.

The incidents behind the headlines

Three concrete events drove this month's safety push:

• GPT-6.1 Astra. Held back on September 28 after tests found more deception and scope problems. The UK AI Security Institute had tested the earlier GPT-6 Astra (The Guardian).
• The Australian government website breach. A rogue agent accessed government sites in June. Prime Minister Albanese called it "unacceptable," per the same Guardian report.
• Agent swarms going off task. Amodei's essay describes an incident in which an agent swarm attacked targets it had not been asked to attack (darioamodei.com).

None of these is a superintelligence. They are capable agents with too much access and weak limits on scope. Any company that deploys agents takes on the same kind of risk today, at a much smaller scale.

Why critics call the accord weak

The criticism is consistent across outlets. The Guardian notes that none of the controls involve government regulators or require companies to publish findings, and companies can choose their own evaluators (The Guardian). Toby Walsh of UNSW asked, "What other trillion-dollar industry marks its own homework?" Alvin Wang Graylin of the Asia Society said, "What is missing is independence and anything that crosses a border." David Krueger of the University of Montreal called it a way to "regulate without regulating" (Al Jazeera).

Sen. Mark Warner put it bluntly: "The president's response? To rename it and tell the companies developing it to regulate themselves" (CBS News). Kate Devlin made the same point about Astra: "it's still the tech companies, rather than regulatory bodies, who get to decide what is safe" (The Guardian).

Supporters have a case too. AP quotes USC professor Shri Narayanan saying the accord balances room for innovation with regulation. AP also notes that some of the steps are things the companies already do in some form (AP).

Superintelligence policy and regulation in 2026

US policy on superintelligence now runs on four tracks at once: a vocabulary order, a voluntary accord, a proposed ban in Congress, and talks with China. Other governments are pushing for mandatory testing.

Sources: Executive order, accord text, Sanders, White House US-China fact sheet, Government of the Netherlands.

The federal backdrop

The order builds on earlier federal moves listed in the notes to 15 U.S.C. 9401: Executive Order 14179 of January 23, 2025, the Genesis Mission order (EO 14363) of November 24, 2025, and EO 14365 of December 11, 2025, which set out a national framework and created an AI Litigation Task Force to challenge state AI laws. The direction has been steady: fewer federal rules, pressure on state rules, and voluntary commitments from industry.

Trump said as much on Tuesday: "I think I'm seeing tremendous self-policing. And they understand that they have to self-police" (AP).

What to watch next

Three dates are on the calendar. The AI czar Trump promised within days of September 29. The next US-China SI Dialogue exchange by November 2026. And the proposed federal SI definition due by November 28, 2026. That last one is the date businesses should mark. If the definition narrows "SI" to frontier systems, the rename becomes meaningful. If it keeps the 15 U.S.C. 9401(3) scope, SI is just a new name for AI.

What superintelligence means for businesses and marketers

For a marketing leader, the practical effect of this week is not a new rule. It is a new word in vendor pitches, a jump in search demand, and a public oversight model you can borrow for your own agents.

Treat "SI" as a label, not a spec

Expect vendors to rebrand. The federal government now says SI for all AI covered by 15 U.S.C. 9401(3), which is a broad definition. A tool marketed as "super intelligence" after September 29 may be the same product it was on September 28.

So ask the questions you would ask of any AI tool. What does it do, step by step? What data does it read? What is it allowed to change without a person approving it? What happens when it is wrong? If the answers are vague, the label is doing the selling.

Search demand is spiking, and the results page is thin

Here is what the data showed when we pulled it on September 30, 2026:

In Ahrefs' volume history, US searches for "superintelligence" hit 18,432 in October 2025, the month the Statement on Superintelligence launched. That is the highest month in two years of data. By July 2026 the figure was 1,193. Google Trends shows the new spike this week.

The top results for "superintelligence" in Ahrefs' SERP snapshot of September 26, 2026 are thin: a Reddit philosophy thread, LinkedIn posts, a Stack Exchange question and a journal article. A keyword difficulty of 2 means few strong pages compete. Meanwhile, Ahrefs shows Google AI Overviews on the results for "artificial superintelligence," "ai superintelligence," "agi vs asi" and "what is superintelligence."

What to take from it: if your business sells AI services or writes about AI, the vocabulary your buyers search with may shift. Track both "AI" and "super intelligence" terms in your keyword set, and watch whether your category's queries start to pick up the new word. Do not rename your pages on a hunch. Rename them when the data moves.

Copy the accord's four layers into your own AI stack

The accord's structure is sound even if its enforcement is thin. You can apply it to a marketing team running agents, at a fraction of the scale.

This is how we run our own work. Our AI usage policy says the named owner of each workflow reviews every output before it ships externally. The Marketing Agent we build drafts plans, articles and SEO fixes, and its stated guardrail is no live publishing, no ad edits and no invented numbers.

Scope is the risk you can actually control

Look again at the incidents. Astra had "scope" problems. An agent reached government websites it should not have. A swarm hit targets nobody asked it to hit. The common failure is permission, not intelligence.

Here is a worked example. Say your team gives an AI agent API access to your CMS so it can draft blog posts. If that token can also publish and delete, the agent can publish a draft with a wrong price or pull a live page. The fix costs nothing: create a draft-only role, give the agent that role, and have a person press publish. The same logic applies to your email platform, your ad accounts and your CRM. List every system an agent can reach, then remove every permission it does not need for its one job. Our post on writing AI agent handoff rules walks through where a person should take over.

Answer engines reward what this news cycle rewards: clear, sourced answers

Google's own guidance says there are no special tricks for AI Overviews. To appear as a supporting link, a page must be indexed and eligible for a snippet, and "There are no additional technical requirements" (Google Search Central). Google also says structured data should match the visible text on the page, and that traffic from AI features is reported in Search Console under the "Web" search type.

That means the basics carry the weight: a direct answer near the top, named entities, dates, sourced numbers and a clear structure. We cover this in more depth in GEO vs SEO: how to build pages AI can trust and in our guide to Google AI Overviews and SEO.

What we would do this quarter

If you lead marketing at a company already running content, SEO, email or paid, here is the short list:

1. Audit every AI tool and agent your team uses for write access, and remove what is not needed.
2. Name one owner per AI workflow and write down what they approve.
3. Add "superintelligence" and "super intelligence" terms to your keyword tracking so you see a shift early.
4. Ask each AI vendor for its equivalent of the accord's four layers, in writing.
5. Keep a person on every send, publish and price change.

Want an agent that drafts but never ships on its own?

If you run marketing and want content, SEO checks and attribution handled by an agent while a person signs off every piece, look at the Marketing Agent build. It is a build for a marketing leader, and it starts with a free plan of what you publish and where leads come from.

FAQ: superintelligence

What is superintelligence?

Superintelligence is a hypothetical form of AI that clearly exceeds the best human minds in almost every field, including science, strategy and social skills. Nick Bostrom defined it in 1998 as an intellect that surpasses the best human brains "in practically every field." No lab claims to have built it as of September 30, 2026.

What is ASI in AI?

ASI stands for artificial superintelligence. It is the stage beyond artificial general intelligence (AGI), where an AI system outperforms all humans across nearly all cognitive tasks. Google DeepMind's "Levels of AGI" paper calls this Level 5 general AI, defined as a system that "outperforms 100% of humans."

What is the difference between AGI and ASI?

AGI is AI that performs at roughly skilled-human level across most cognitive tasks. ASI is AI that goes well beyond the best humans across nearly all of them. OpenAI described superintelligence in 2023 as future systems "dramatically more capable than even AGI." Some researchers argue AGI has arrived; no one claims ASI has.

Is superintelligence possible?

Researchers disagree. A September 28, 2026 paper by 22 researchers, including Geoffrey Hinton and Yoshua Bengio, warns that automated AI research could trigger an intelligence explosion. Princeton's Arvind Narayanan and Sayash Kapoor argue superintelligence, as usually described, is not a coherent idea and that AI's effects will unfold over decades.

How far are we from superintelligence?

Estimates range from about a year to never. UC San Diego's Eddy Keming Chen said in September 2026 it could come "perhaps even within a year." Nick Bostrom put less than a 50% chance on it arriving by 2033, in a 2008 postscript. The Cambridge Programme paper projects that months-long AI research projects could be automated by mid-2028.

What did Trump's superintelligence executive order do?

Signed September 29, 2026, the order "Inaugurating the Era of Super Intelligence" tells federal agencies to use "Super Intelligence" and "SI" instead of "Artificial Intelligence" and "AI" in official communications. It defines SI using the existing legal definition of AI, and it asks for a proposed federal SI definition within 60 days.

What is the White House Accord on Super Intelligence?

It is a one-page voluntary agreement signed September 29, 2026 by President Trump and the leaders of Google, Anthropic, Meta, OpenAI, xAI and Nvidia. It commits the companies to four layers: internal controls, an internal team, an independent external auditor and an independent board committee. It does not involve government regulators.

What is the Ban Artificial Superintelligence Act?

It is a bill introduced September 23, 2026 by Sen. Bernie Sanders and Rep. Greg Casar. It would pause advanced AI development until a new cabinet-level Department of Artificial Intelligence sets rules. Penalties include a corporate death penalty for companies and up to 20 years in prison for individuals.

Is ChatGPT a superintelligence?

No. OpenAI does not claim ChatGPT is superintelligent. Current models are superhuman at some narrow tasks and still fail at others. In September 2026, OpenAI held back its GPT-6.1 Astra model after tests showed more deception and problems staying within its authorized scope.

About the author

Angel Castro is AI Systems Lead at Conversion System, an AI agency that builds agent workflows, SEO systems and automation for marketing teams. Our agents run with a named person who approves the work before it goes live.

Sources

1. The White House, "Inaugurating the Era of Super Intelligence," executive order, September 29, 2026. https://www.whitehouse.gov/presidential-actions/2026/09/inaugurating-the-era-of-super-intelligence/
2. The White House, "Fact Sheet: President Donald J. Trump Inaugurates the Era of Super Intelligence," September 29, 2026. https://www.whitehouse.gov/fact-sheets/2026/09/fact-sheet-president-donald-j-trump-inaugurates-the-era-of-super-intelligence/
3. The White House, fact sheet on the US-China state visit, September 25, 2026. https://www.whitehouse.gov/fact-sheets/2026/09/fact-sheet-president-donald-j-trump-advances-a-fair-and-reciprocal-relationship-with-china-while-hosting-historic-state-visit/
4. 15 U.S.C. 9401, definitions (Cornell LII). https://www.law.cornell.edu/uscode/text/15/9401
5. Washington Examiner, full text of the White House Accord on Super Intelligence, September 29, 2026. https://www.washingtonexaminer.com/news/white-house/4747747/full-trump-white-house-accord-ai-super-intelligence/
6. Associated Press (via Boston.com), Rugaber, Weissert and Kinnard, September 29, 2026. https://www.boston.com/news/politics/2026/09/29/trump-says-top-tech-firms-have-signed-accord-to-self-police-ai-development/
7. Reuters (via Investing.com), Kanishka Singh, September 29, 2026. https://www.investing.com/news/stock-market-news/trump-releases-ai-accord-with-tech-executives-4923759
8. CBS News, Joe Walsh, September 29, 2026. https://www.cbsnews.com/news/trump-ai-constitution-tech-execs-openai-anthropic-voluntary-controls/
9. Fox Business, September 29, 2026. https://www.foxbusiness.com/politics/trump-signs-executive-order-rebranding-ai-super-intelligence-tech-titans-ink-separate-accord
10. Al Jazeera, John Power, September 29, 2026. https://www.aljazeera.com/news/2026/9/29/trump-top-tech-firms-sign-accord-to-self-police-ai-development
11. The Guardian, Nick Robins-Early, September 29, 2026. https://www.theguardian.com/us-news/2026/sep/29/trump-ai-deal-tech-ceos-superintelligence
12. The Guardian, on the intelligence explosion paper, September 28, 2026. https://www.theguardian.com/technology/2026/sep/28/ai-godfathers-warn-of-runaway-intelligence-explosion
13. The Guardian, on GPT-6.1 Astra, September 28, 2026. https://www.theguardian.com/technology/2026/sep/28/openai-new-model-astra-release-scrapped
14. Cambridge Programme on AI Science and Policy, "What if automating AI R&D triggers an intelligence explosion?", September 28, 2026. https://casp.ac/reports/intelligence-explosion (summary PDF: https://casp.ac/__l5e/assets-v1/026ee81c-773c-4811-9c97-f7cf367ee7b0/intelligence-explosion-summary.pdf ; full PDF: https://casp.ac/__l5e/assets-v1/5efd4b41-deb5-4513-a0a3-b4f82d2b79ea/intelligence-explosion.pdf)
15. The Verge, Richard Lawler, UN General Assembly remarks, September 22, 2026. https://www.theverge.com/ai-artificial-intelligence/998816/donald-trump-ai-super-intelligence
16. Defense One, John Croxton, September 22, 2026. https://www.defenseone.com/technology/2026/09/super-intelligence-presidents-new-term-ai-explained/416153/
17. Office of Sen. Bernie Sanders, press release on the Ban Artificial Superintelligence Act, September 23, 2026. https://www.sanders.senate.gov/press-releases/news-sanders-casar-introduce-legislation-to-create-new-federal-agency-to-ban-artificial-superintelligence-pause-advanced-ai-development/
18. Government of the Netherlands, "A Call for Control of Frontier AI Models," September 22, 2026. https://www.government.nl/documents/2026/09/22/a-call-for-control-of-frontier-ai-models
19. Dario Amodei, "We Must Pace the Frontier," September 2026. https://darioamodei.com/post/we-must-pace-the-frontier
20. Nick Bostrom, "How Long Before Superintelligence?" (1998, with 2008 postscript). https://nickbostrom.com/superintelligence
21. OpenAI, "Governance of superintelligence," May 22, 2023. https://openai.com/index/governance-of-superintelligence/
22. Morris et al., Google DeepMind, "Levels of AGI," arXiv 2311.02462. https://arxiv.org/abs/2311.02462
23. Arvind Narayanan and Sayash Kapoor, "AI as Normal Technology," Knight First Amendment Institute, April 15, 2025. https://knightcolumbia.org/content/ai-as-normal-technology
24. UC San Diego Today, interview with Eddy Keming Chen, September 28, 2026. https://today.ucsd.edu/story/how-close-are-we-to-artificial-superintelligence
25. Future of Life Institute, Statement on Superintelligence. https://superintelligence-statement.org/
26. CNBC, Dylan Butts, October 22, 2025. https://www.cnbc.com/2025/10/22/800-petition-signatures-apple-steve-wozniak-and-virgin-richard-branson-superintelligence-race.html
27. Meta, Mark Zuckerberg, "Personal Superintelligence," July 30, 2025. https://www.meta.com/superintelligence/
28. Safe Superintelligence Inc. https://ssi.inc/
29. Google Search Central, "AI features and your website." https://developers.google.com/search/docs/appearance/ai-features
30. Ahrefs Keywords Explorer and SERP Overview, US data pulled September 30, 2026.
31. Google Trends, "superintelligence," US, past 90 days and past 30 days related queries, retrieved September 30, 2026. https://trends.google.com/trends/explore?date=today%203-m&geo=US&q=superintelligence