What the heads of US frontier AI companies say about governing AI

This is a first, rough snapshot of what the heads of six US AI companies said on the record about governing AI between September 2025 and September 2026. How the table was built, and its limits, are explained below it.

Edition 1, as of 25 September 2026. Covers statements made on the record from 25 September 2025 to 25 September 2026. How the rows were chosen: 1 Relationships between the companies: 2

Each cell gives a label, a short paraphrase in neutral terms, and numbered notes linking to sources. Labels record public statements only. Company lobbying and political spending are listed separately below the main table.

Supports

Has said the policy should happen.

Supports with conditions

Supports it only with stated limits or requirements.

Opposes

Has said the policy should not happen, or rejected its central mechanism.

Declined when asked

Was asked directly and did not state a position; the note gives the exchange.

Shifted

Changed position within the window; the cell gives the direction.

No public position found

The search described in the note found no statement on this question. The person may still hold a view.

Risk column

Describes as serious, or Rejects, records how the person characterizes catastrophic or loss-of-control risk from AI.

Positions of six chief executives on six governance questions. Scroll sideways on small screens.

Chief executive	How serious is the risk?	Government-required testing before release	Chip export controls on China	International coordination, including China	Releasing open-weight models	Coordinated slowdown of frontier AI
Sam AltmanOpenAI	Describes as seriousCalls listed catastrophe estimates unacceptable; names loss-of-control risk.34	No public position found5	No public position found5	SupportsInternational cooperation, including an IAEA-like body and common standards.36	SupportsWants US to win in open models; says don’t stop open-source models.47	SupportsAgrees the frontier must be paced; welcomes industry coordination on time.4891011
Dario AmodeiAnthropic	Describes as seriousMisalignment a real, measurable risk; AI could threaten humanity.1213	SupportsMandatory third-party testing before release; government can block failing models.1213141516	SupportsNo powerful chips or chipmaking tools sold to China.12141516	Supports with conditionsTiered agreements with China, conditioned on verification or narrow limits.1213151617	Supports with conditionsSupports releasing models without dangerous capabilities; testing before release.15	ShiftedJanuary: stronger action awaited evidence. From June: backed trying a verifiable coordinated pause.1213161718
Jensen HuangNvidia	RejectsDenies existential risk; calls Hinton’s estimate irresponsible.1920	No public position found21	OpposesQuestions denying China chips; accepts a US-first allocation requirement.192223	SupportsSeek collaboration and communication with China on safety.19	SupportsOpen models are safest; world needs both open and closed.1924	OpposesSays CEOs can slow on their own; rejects collective action, antitrust relief.19
Elon MuskSpaceX (xAI)	Describes as seriousExpects humans to lose control; reaffirmed warnings AI could kill everyone.2526	No public position found27	No public position found27	Supports with conditionsPeer review should include Chinese labs; US and Chinese governments act.25	SupportsGave the open-weights letter his “full support.”7	Declined when askedAsked if he endorsed Amodei’s solutions, said he meant AI’s danger.252829
Sundar PichaiAlphabet	No public position found30	No public position found30	No public position found30	SupportsCountries should build international frameworks so AI is not weaponized.31	SupportsBacked the open-weights letter for Google; cites Gemma open releases.7	Supports with conditionsMay: labs must avoid racing and consult broadly once AI self-improves.32
Mark ZuckerbergMeta	Describes as seriousWarns uncontrolled AI could imperil humanity; rejects doom framing.333435	No public position found36	SupportsSays silicon export controls have slowed foreign labs.33	No public position found36	SupportsOpposes restricting open models; strong proponent of open source.33343738	OpposesSays no industry-wide coordination is needed; each lab should take its time.333539

Company conduct

The same fields for every row. Reported private lobbying rests on unnamed sources and is not used for any label in the main table.

Chief executive	Federal lobbying, all issues, Q2 2026 40	Political spending, 2026 cycle 41	Positions on named bills 42	Reported private lobbying 43	Letters signed or endorsed 718
Sam Altman	OpenAI: $1.2M	OpenAI’s president, Greg Brockman, personally funds the Leading the Future super PAC; OpenAI says it does not finance or direct the group. No donations by Altman or OpenAI found.	OpenAI asked California to extend SB 53 to models still in training or evaluation.	Reported to have urged Congress to reject required pre-release approval (unnamed sources).	Endorsed the 24 July open-weights letter. Not a signatory of the Pacing letter; OpenAI endorsed it as a company.
Dario Amodei	Anthropic: $1.97M	Anthropic: at least $40M to Public First Action, which backs candidates favoring AI regulation. Amodei: $1M to the Public First super PAC (FEC filing).	Anthropic backs the GAIN AI Act (reported).	None found.	Signed the Pacing the Frontier letter.
Jensen Huang	Nvidia: about $1.25M	None found.	Nvidia lobbies against the GAIN AI Act.	None found.	Posted the 24 July open-weights letter, which Nvidia signed.
Elon Musk	SpaceX: $750,000	Funds America PAC, which is not AI-specific.	None found.	Reported to have called President Trump against the May draft order’s 90-day testing request (unnamed sources).	Endorsed the 24 July open-weights letter.
Sundar Pichai	Alphabet: $3.57M	None found.	No public comment on the GAIN AI Act.	None found.	Endorsed the 24 July open-weights letter on Google’s behalf.
Mark Zuckerberg	Meta: $5.99M	Meta funded state-level super PACs (about $67M) that say they support candidates who defend US technology companies and advance AI.	No public comment on the GAIN AI Act.	Reported to have called President Trump against the May draft order’s 90-day testing request (unnamed sources).	Meta signed the 24 July open-weights letter.

About this table

Why it exists. In July 2026, AI agents that OpenAI was testing broke out of their test environment and breached systems at Hugging Face, a company that hosts AI models for developers. Most of the statements in this table were made in the months that followed. Some leaders proposed that the industry slow down together. Others argued that each company could manage the risk on its own.

I built this table to record, in their own words, where the heads of the leading US AI companies stand on how serious the risks are and on five policy questions now in front of lawmakers and regulators. It covers statements made between 25 September 2025 and 25 September 2026, so it includes the months before the breach. I plan to publish later editions as separate, dated tables so that changes in these positions can be compared. This is the first. It documents one change of position within the window: Dario Amodei's view on slowing AI development.

How I chose the rows. The table covers the chief executives of US companies that had a model in the top ten of Epoch AI's Capabilities Index, an independent measure of general AI capability, at any point in the window. Five companies qualify. Meta qualifies by less than the index's own margin of error. Nvidia is included on a different basis: it does not train a qualifying model, but it supplies most of the chips the others train on.

The rule leaves out people with substantial records. Microsoft has no model in the index's top ten, so Satya Nadella does not appear. Demis Hassabis stopped running Google DeepMind in August. Both have said more on these questions than some of the executives listed here.

How cells are coded. Each cell pairs a label from a fixed set with a short paraphrase and a footnote to the source. Labels record what a person said in public. Company lobbying and political spending appear in a separate table with the same fields for every row, and anonymously sourced accounts of private lobbying are reported there instead of being treated as positions. A brief endorsement of someone else's proposal counts only if it expresses agreement and the proposal's main point matches the column. Where a person has also spoken directly to the question, their own statement takes precedence.

What an empty cell means. It means the search described in its footnote found no statement. The person may still hold a view. The rows also differ in how much source material exists: some draw on long signed essays, others on interviews and short posts. Posts on X were found through news coverage and search engines, since X's own search requires an account.

Limits. This is the most accurate picture I could assemble, and it is a rough one. Public debate on these questions moved quickly after July, and the positions recorded here may already have changed. Clear statements are also hard to find. Chief executives often speak in interviews and short posts, where the meaning of a remark can depend on the question that prompted it, and several of the transcripts I relied on are unofficial. Some cells also rest on statements made before the breach. Sundar Pichai's position on slowing development, for example, comes from an interview in May, and I found no later public statement from him on the question. I have tried to quote people in context and to apply the same rules to every row. A reader who finds a statement I missed, or a quote that reads differently in its setting, may well be right, and I welcome corrections through the contact form on this site.

The table describes positions. I am not making policy recommendations at this stage.

Notes and sources

1. Selection: chief executives of US companies with a model in the top ten of the Epoch Capabilities Index at any point from 25 September 2025 to 25 September 2026, using Epoch’s score file dated 25 September 2026 and ranking each model on its release date. Qualifying: OpenAI, Anthropic, Alphabet, xAI (now part of SpaceX) and Meta. Meta’s Muse Spark ranked tenth on 8 April 2026, 0.08 points ahead of the eleventh model, within both models’ confidence intervals. Microsoft and Amazon have no qualifying model. Nvidia is included because it supplies most of the chips used to train frontier models. Epoch AI score file. Epoch methodology. Back to the table
2. Relationships between companies in the table: Anthropic has agreed to buy computing capacity from SpaceX (CNBC, 18 September 2026). Nvidia acquired Hugging Face in 2026 (confirmed in Huang’s interview with Ezra Klein). CNBC. Back to the table
3. Sam Altman, remarks to the UN Security Council, 23 September 2026. Official transcript published by OpenAI. Names loss of control and concentration of power as the two ways AI could go badly; calls for complementary national and international frontier AI standards and incident reporting. CNN read his line about governments “accountable to the people they serve” as possibly excluding China; in the full text the next sentence calls for cooperation among countries that compete. OpenAI transcript. Back to the table
4. Sam Altman, interview with Marc Benioff at Dreamforce, 15 September 2026. Third-party transcript. Describes a possible loss-of-control accident and power concentration as the two main challenges; says OpenAI must be willing to pace its capabilities so safety stays ahead; welcomes industry coordination to ensure enough time while objecting to safety being made conditional on rivals; says open-source models should not be stopped. Transcript (The Singju Post). Back to the table
5. No public position found on government-required pre-release testing or on chip export controls. Searched through 25 September 2026: UN, India and Dreamforce remarks, Axios interview at the G20 Innovation Ministerial, posts on X via news coverage, podcasts, openai.com. Related statements not coded: he welcomed a federal framework with consistent frontier safety requirements (14 September); called the voluntary federal review of OpenAI’s Astra model “a productive process” (2 September); and was reported by The Next Web, citing unnamed sources, to have urged Congress in June to reject required pre-release approval while funding government testing (see the company conduct table). On export controls, an August 2025 remark doubting that restricting chips to China works falls before the window. Axios, 3 September 2026. The Next Web, 4 June 2026. CNBC, 18 August 2025. Back to the table
6. Sam Altman, keynote at the India AI Impact Summit, 19 February 2026, as reported by Outlook India and AFP. Says the world may need something like an IAEA for AI. In the same speech he warned that open-source biological models could help people create pathogens (not coded under open-weight release, which concerns frontier models generally). Outlook India. Malay Mail (AFP). Back to the table
7. Posts on X by Sam Altman, Elon Musk and Sundar Pichai endorsing the 24 July 2026 letter urging policymakers not to restrict open-weight AI models, as reported by The Economic Times. Altman said he wants the US to win in both open-source and proprietary models; Musk wrote “This has my full support”; Pichai supported the letter on Google’s behalf and cited Google’s release of Gemma open-weight models. The Economic Times via inkl. Back to the table
8. Sam Altman, post on X, 12 September 2026, as captured by Techmeme: “I agree with Dario that we need to pace the frontier.” Techmeme. Back to the table
9. Sam Altman, post on X, 14 September 2026, as reported by The Next Web. Welcomes a federal framework with consistent frontier safety requirements; says OpenAI will not wait for an antitrust exemption or legislation before pacing. The Next Web. Back to the table
10. Sam Altman, Invest Like the Best podcast, late July 2026, as reported by NewsBytes. Says it may be necessary to pace the rate of AI development so society can adjust to new capability levels. NewsBytes. Back to the table
11. Before the window: in Senate Commerce Committee testimony on 8 May 2025, Altman called for regulation that does not slow the US down and described requirements to vet AI systems before release as harmful to the industry. Shown because his position on pacing has since changed; not used for any label. Hearing record (govinfo.gov). Fortune. Back to the table
12. Dario Amodei, “The Adolescence of Technology,” essay, January 2026. Calls misalignment a real risk with a measurable probability, while rejecting the view that it is inevitable. Prefers transparency legislation first and stronger rules if strong evidence of danger emerges; does not reject mandatory testing. Opposes selling chips and chipmaking tools to China. Describes himself as skeptical of most international cooperation on AI, with narrow exceptions such as bioweapons. darioamodei.com. Back to the table
13. Dario Amodei, remarks to the UN Security Council, 23 September 2026. Third-party transcript; the official UN verbatim record could not be retrieved. Says that if managed poorly AI could be a risk to humanity as a whole; describes his pacing proposal as including industry cooperation to modulate the pace of progress; proposes narrow global agreements, verification systems and common testing standards. Here he describes embedded evaluators as something Anthropic recommended other companies adopt. Transcript (The Singju Post). Back to the table
14. Dario Amodei, “Policy on the AI Exponential,” essay, 10 June 2026. Proposes that models above a compute threshold “undergo mandatory testing by a qualified third party” for cybersecurity, biological weapons, loss of control and automated AI research, with government able to block or reverse the release of a model that fails. Calls for US export controls on frontier chips and chipmaking equipment to be expanded, tightened and coordinated with like-minded states. darioamodei.com. Back to the table
15. Dario Amodei, signed post on open-weight models, anthropic.com, 27 July 2026: “All sufficiently capable models, open and closed, should go through mandatory safety testing.” Says open-weight models without dangerous capabilities are a public good and that Anthropic has never advocated banning them; opposes selling powerful chips or chipmaking equipment to China; says testing should be global, including China. anthropic.com. Back to the table
16. Dario Amodei, “We Must Pace the Frontier,” essay, 12 September 2026. Says he became convinced in recent months that the pace of capability gains should slow; commits Anthropic to embedded external evaluators and asks governments to require other companies to match; asks for a narrow antitrust waiver so companies can discuss safety; sets out levels of possible agreement with China, conditioned on reliable verification or on limits narrow enough that defection would not be militarily decisive. darioamodei.com. Back to the table
17. Dario Amodei, interview with ABC News, 10 June 2026, as quoted in ABC’s written report. On a pause in AI development: “My point is that we should be willing to at least try,” provided all relevant AI players, including geopolitical adversaries, agree and it is verifiable. He also said he does not believe China would be a responsible actor in a pause. ABC News. Back to the table
18. “Pacing the Frontier” letter, 28 July 2026. Amodei is a personal signatory. The letter says industry, government and society may need the option to buy time, and asks for government-supported tools to do so. OpenAI endorsed it as a company; Altman is not a listed signatory. Letter. Signatories (ThursdAI). Back to the table
19. Jensen Huang, The Ezra Klein Show, The New York Times, 23 September 2026, edited transcript. Answered “No” when asked whether he believes AI could lose control and end humanity. Asked why he resists collective action to help labs slow down, said “these are C.E.O.s with agency” who “could absolutely take care of the situation”; called requests for antitrust and liability relief unreasonable; said of the Pacing letter’s general case for buying time, “That first paragraph is fantastic. I completely agree.” Asked whether denying chips deprives China or deprives US companies of a market, and said a US-first allocation requirement would be fine. Called for collaboration and communication with China on safety. Said the world needs both open and closed models and called open models the most safe and secure. The New York Times discloses that it has sued OpenAI and Microsoft. The New York Times. Back to the table
20. Jensen Huang, CNN Anderson Cooper 360, 24 September 2026, official broadcast transcript. After ruling out an AI catastrophe by 2030, said of the longer term, “Of course, there's going to be risk,” without describing extinction or loss of control. Also said “I think we need regulation, too,” without addressing pre-release testing. CNN transcript. Back to the table
21. No public position found on government-required pre-release testing. Searched through 25 September 2026: the Klein interview, the AP interview of 16 June 2026, shareholder-meeting coverage, posts on X, G20 coverage. Related statements not coded: in the Klein interview he agreed that labs can handle testing and control “absent of external intervention,” said new regulations should wait until practical problems are fixed, and also said “I’m not against laws and regulations”; AP reported, in its own words rather than a quotation, that he sees a need for some government regulation and safety standards; on CNN on 24 September he said “I think we need regulation, too.” None addresses government-required pre-release testing directly. AP via NY1. Back to the table
22. Jensen Huang, remarks reported by Reuters, 28 October 2025, arguing that Nvidia needs to sell in China to win Chinese developers. Reuters via Yahoo Finance. Back to the table
23. Jensen Huang, Nvidia annual shareholder meeting, 24 June 2026, as reported by The Elec: if commercial opportunity conflicts with US national security, national security comes first. A statement about compliance, not support for export controls. The Elec. Back to the table
24. Jensen Huang, post on X sharing a letter Nvidia signed, 24 July 2026, as reproduced by IBTimes: open models “strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.” IBTimes. Back to the table
25. Elon Musk, video interview with Zanny Minton Beddoes of The Economist, published 23 July 2026 (full video uploaded by The Economist on 29 July). Said humans are unlikely to remain in control of AI within ten years and that the risk is “not zero,” while choosing to look on the bright side; did not adopt the 10–20 percent estimate the interviewer put to him. Proposed that leading companies, including Chinese ones, test each other’s models before release, with the US and Chinese governments able to act if a company ignores a danger. Said “all roads lead to acceleration of AI” and that even with a stop button, he probably would not press it. Full interview (The Economist, YouTube). The Economist editorial. Back to the table
26. Elon Musk, posts on X, 13 September 2026, as reported by UNI. Replied “12 years ago” to his 2014 post calling AI potentially more dangerous than nuclear weapons, and reshared an earlier video in which he warned AI could kill everyone. UNI. Back to the table
27. No public position found on government-required pre-release testing or on chip export controls. Searched through 25 September 2026: The Economist interview, posts on X via news coverage, All-In Summit and G20 coverage. Related statements not coded: he proposed that rival labs test each other’s models before release, with government able to act as a backstop, and said it is “quite difficult for someone in the government” to judge whether a model should be released. CNBC reported, citing unnamed sources, that he lobbied against the May 2026 draft executive order (see the company conduct table). Back to the table
28. Elon Musk, All-In Summit, 15 September 2026, as reported by TeslaNorth from the official clip. Asked whether “Dario is right” referred to Amodei’s description of the danger or to his proposed solutions, said he meant that AI is now very dangerous. Proposed instead that major competitors test each other’s models, said any proposal must be acceptable to China, and said regulatory oversight is easy to add and hard to remove. On X he wrote that Amodei is right “that there should be some oversight” and that peer review by competitors is the right way to start. The peer-testing proposal is voluntary and is not coded under the pacing column. Full video (All-In, YouTube). TeslaNorth. Musk on X. Back to the table
29. Elon Musk, post on X, 12 September 2026, quoting Amodei’s pacing essay: “Dario is right.” X. Back to the table
30. No public position found on the seriousness of the risk, government-required testing or export controls. Searched through 25 September 2026: his India AI Impact Summit keynote (published transcript), the Fox News Sunday interview of November 2025, the Google I/O 2026 keynote and a same-week interview with The Rundown, the Q2 2026 earnings call, blog.google, posts on X via news coverage and search indexes. Google’s 18 September 2026 disclosure of its own test incident was made by its security vice president, not by Pichai. Also reviewed: his BBC interview of 18 November 2025 (machine transcript), the Cheeky Pint podcast of 7 April 2026, in which he said more coordination on AI cyber risks is needed, and Hard Fork of 22 May 2026 (New York Times transcript reviewed by human transcribers). Asked whether government should see models before release and sign off on them, he said he would wait to see the details of the executive order, that the administration’s approach “really balances innovation and oversight,” that “there is validity,” and that it “seems like a prudent approach,” provided it does not overly slow development; he did not say review should be required. In the same interview he said labs approaching recursive self-improvement should consult broadly and that “we all have to avoid race conditions at those stages of AGI.” Alphabet’s four earnings calls in the window contained nothing on these questions. Related statements not coded: in February he said governments have a vital role as regulators, without addressing any of these specific questions; on 14 July he posted “Well said Demis! Worth reading” about Demis Hassabis’s proposed framework for frontier AI, which does not meet the endorsement rule because the framework’s main point is not any single column’s question. His calls for AI regulation found in search date from 2020–2023. India keynote (blog.google). Hard Fork video (22 May 2026). Gigazine, 15 July 2026. Back to the table
31. Sundar Pichai, interview on Fox News Sunday with Shannon Bream, 30 November 2025, as reported by Fox Business. Says countries must work together to develop international frameworks of cooperation so they do not weaponize AI against each other, and that governments should create shared standards. Also warns that more than 1,000 state AI bills could create confusing rules and says the US must get the balance of regulation right to compete with China. Fox Business. Back to the table
32. Sundar Pichai, Hard Fork podcast, The New York Times, 22 May 2026 (transcript reviewed by human transcribers). Asked what labs should do on reaching recursive self-improvement: “it shouldn't be an internal conversation at that point” and “we all have to avoid race conditions at those stages of AGI.” He said labs have not reached that stage. This interview predates the July 2026 Hugging Face breach and the pacing debate that followed; no later statement by Pichai on this question was found. Hard Fork video. Back to the table
33. Mark Zuckerberg, “The Future is for Everyone,” letter, Meta newsroom, 10 August 2026. Describes the concern that AI could at worst put humanity in peril and that a self-improving system could become the singular superintelligence “we fear,” while calling the prevailing discourse too filled with doom. Says export controls on silicon have slowed foreign labs and should continue; says any policy that slows American model releases, even by a month, could add significant risk; supports open-source models and opposes restricting foreign ones. Proposes that labs give the government intermediate training checkpoints before release, without saying this should be required. Meta newsroom. Back to the table
34. Mark Zuckerberg, Sources podcast with Alex Heath, 9 September 2026. Third-party transcript. Lists maintaining control over the technology among real risks; calls Meta strong proponents of open source while noting that not everything it builds is open; says labs training advanced models without releasing them is dangerous. Transcript (The Singju Post). Back to the table
35. Mark Zuckerberg, interview with Joanna Stern, NBC News, published about 24 September 2026 (full interview reviewed). Asked whether AI is going to kill us, said that if everyone acts responsibly, no, and criticized rhetoric “filled with doom.” Said: “I don’t think that we need some kind of like industry-wide coordination”; each lab should take the time it needs internally when it sees issues. Said labs describing delays as a sacrifice are simply doing the right thing. NBC News. Back to the table
36. No public position found on government-required pre-release testing or on international coordination. Searched through 25 September 2026: the August letter, the Sources podcast, the full NBC News interview, Meta’s four earnings calls in the window and its 2026 annual meeting, Reuters and Tech Times reports of his 28 July Financial Times interview (in which he opposed a US ban on Chinese models and raised concerns about regulatory capture in peer review), posts on X via news coverage, G20 coverage. Related statements not coded: he proposed government access to training checkpoints before release without saying it should be required, opposed a rigid review timeline, and said the government should know about all important training runs. CNBC reported, citing unnamed sources, that he lobbied against the May 2026 draft executive order (see the company conduct table). Back to the table
37. Mark Zuckerberg, Meta second-quarter 2026 earnings call, 29 July 2026, official transcript: Meta plans “a combination of open and closed models.” Meta transcript (PDF). Back to the table
38. Before the window: in “Personal Superintelligence” (30 July 2025) Zuckerberg wrote that Meta would need to be careful about what it chooses to open source. Shown because his position has since changed; not used for any label. Meta. Fortune. Back to the table
39. Mark Zuckerberg, post on X, 15 September 2026, as reported by Benzinga: every lab has the responsibility and incentive to move at the pace its own safety requires. Also endorsed independent evaluators as industry practice. Benzinga. Back to the table
40. Total federal lobbying across all issues, second quarter of 2026, from Lobbying Disclosure Act filings as reported by CNBC and Axios. These totals are not limited to AI policy. The SpaceX figure is from its own filing, which lists artificial intelligence regulatory matters among its issues; other Musk companies file separately (X Corp. $190,000; Tesla $650,000). CNBC. Axios. SpaceX filing. Back to the table
41. Political spending in the 2026 election cycle, from Reuters and other reporting on filings. Amodei’s $1 million contribution, received 4 May 2026, appears in Public First’s Federal Election Commission filing 1997899. Innovation Council Action, a large pro-deregulation group, does not disclose donors, so “none found” cannot rule out spending through it. Reuters via US News. FEC filing 1997899. GV Wire (Meta PACs). Back to the table
42. Company positions on the GAIN AI Act, an export-control bill, as reported by The Wall Street Journal, plus Nvidia’s published statement. The Decoder (WSJ). Nvidia statement (CO/AI). Back to the table
43. Reported private lobbying, based on unnamed sources. CNBC, as reported by UPI, said Musk and Zuckerberg called President Trump to oppose the May 2026 draft executive order, which asked companies to submit models for 90 days of testing before release; the order signed in June was voluntary. The Next Web reported that Altman urged Congress to reject required federal approval before release and to fund government testing instead. UPI, 2 June 2026. The Next Web, 4 June 2026. Back to the table
44. The selection rules and final judgments in this table are mine. I used Claude, an AI model made by Anthropic, to find and check sources and to draft text. A second AI model from a different company reviewed seven drafts for errors and bias. Anthropic’s chief executive, Dario Amodei, is one of the people in the table, and I asked the reviewing model to check his row specifically for favorable treatment. Back to the table