Evidence as of September 29, 2026. Every factual statement below comes from a retained copy of a dated source in this story's evidence packet, or from the video record page. The packet records dates to the day only, and is explicit about its own limits: attribution, source identity and located excerpts are checked, while "[t]ruth, rights clearance and public release require editorial review." Nothing here is a forecast.
September 2026 produced an unusual kind of disagreement in the United States. Four prominent voices addressed advanced AI in formal settings — a House leadership press conference, the UN General Assembly, the UN Security Council and a senator's own channel. They did not disagree about whether the technology matters. They disagreed about who, if anyone, is supposed to slow it down.
Watch the video record. The video carries the original recordings with their own audio; this is the written companion, with locators and dates.
The shared premise
The agreement is easy to miss because it is never stated as agreement. On September 14 the Congressional Record carries Senator Schumer saying that "signs about AI's rapid pace have never been clearer," and that he was "calling for an all-Senators classified briefing" (pages S4640–S4641). On September 22 it carries Senator Kelly, page S4848, ending a passage on AI with the sentence "And nobody is steering the ship." The same month, the companies building the systems published their own control programmes: Anthropic "announcing Enterprise Frontier Safeguards (EFS)" on September 1 and "introducing the Life Sciences Verification Program (LSVP)" on September 17.
Different speakers, different interests, one shared assumption: this is a steering problem. The four positions below answer who steers. Their ordering, from least to most government control, and the one-line labels are Comuvia's arrangement of the record, not a claim any of the four made about where they sit.
| Position (Comuvia's label) | Who holds it | Dated evidence in the packet |
|---|---|---|
| Industry can regulate itself | Speaker Mike Johnson | C-SPAN clip, September 15 press remarks |
| Encourage it nationally, reject a global scheme | President Trump | UN address, September 22; Executive Order 14409; the September 29 accord |
| Standards, including ones we help write | The chief executives of OpenAI and Anthropic | UN Security Council briefing, September 23 |
| Stop until there are rules | Senator Bernie Sanders | Official Senate office channel video, September 3 |
1. The industry can regulate itself
The strongest version of this position in the September record is a single phrase. A C-SPAN clip of the Speaker's September 15 press remarks is titled "Johnson on AI industry: They can self-regulate" — classified in the packet as a statement attributed through a clip title. The Speaker's own channel retained the full 25-minute House GOP press conference it is cut from, but the fuller remark was checked outside the packet, so that phrase is what we carry as verbatim.
2. Encourage it nationally, reject a global scheme
This position has the most retained text behind it. The White House release of the September 22 UN General Assembly address quotes the President: "The United States totally rejects any attempt to construct a globalist scheme of control for the Artificial Intelligence being spoken of so much now." The domestic direction was already in force: Executive Order 14409, "Promoting Advanced Artificial Intelligence Innovation and Security," signed June 2 and published June 5, 2026, uses the phrase "overly burdensome regulation."
September 29 added a mechanism. Four reports in the packet describe an accord signed with technology leaders at the White House. The Associated Press wire report, carried by KPBS, is headlined "Trump says top tech firms have signed accord to 'self-police' AI development"; its text contains "morally binding," "robust internal controls," "independent external auditor," the line "Over time, it may make sense to codify these steps into laws and regulations," and the line "I will never stifle the growth of a technology." USA TODAY, via Yahoo News, adds "independent committee of the board of directors" and "almost like a constitution." Al Jazeera carries "tremendous self-regulation" and names the document the "Joint Commitment on Frontier Responsibilities." CNBC's report contains "self-policing." The accord itself was posted as an image, so its wording reaches the record only through these reports.
A White House presidential action page of the same date is titled "Inaugurating The Era Of Super Intelligence." The packet locates two strings inside it — a citation to "section 9401(3) of title 15, United States Code," and the phrase "Within 60 days" — and nothing more, so we do not characterise what the order directs.
3. Standards, including ones the builders help write
On September 23 the chief executives of OpenAI and Anthropic briefed the UN Security Council. The packet retains the C-SPAN upload of the UN feed — the full 2-hour-29-minute meeting — but carries no verbatim quotation from either executive; their statements are held as local transcriptions outside it. So the position attributed to them here, that the builders asked for standards, is the video record page's own one-line paraphrase, labelled as such on the card. For their exact words, go to the UN feed.
What the packet does hold from the builders is their own September product and safeguard announcements, recorded as vendor claims and not tested: EFS, LSVP, and OpenAI's launch of GPT-6 Astra, "the most intelligent and aligned model in the world."
4. Stop until there are rules
Senator Sanders published a seven-minute statement on his official Senate office channel on September 3 under the title "Pause AI Development NOW." As with the Speaker and the two executives, the passages inside the video were transcribed locally and are not carried in this packet; the title is the verbatim evidence retained here, and it is unambiguous.
The two open conflicts
Two of these positions collide directly. The Speaker's "they can self-regulate" and the Senator's "pause AI development now" are not points on a spectrum of implementation detail; they are opposite answers to the same question, twelve days apart.
The second conflict runs inside the President's position. On September 22 the record shows a total rejection of a "globalist scheme of control." On September 29 it shows an accord whose press description includes an "independent external auditor," an "independent committee of the board of directors," and the possibility that it "may make sense to codify these steps into laws and regulations." Senator Schumer flagged the first half on September 22: "Trump said he will oppose—oppose—any global effort to regulate AI." Whether voluntary national self-policing with an external auditor differs from a global scheme of control, or is the same thing at a different scale, is the live question. The record does not settle it.
What the rest of the record adds
Five bills were in play, status captured September 29: S. 5518, an "Artificial Intelligence Horizon Fund," with the Committee on Finance; S. 5541, a "Cybersecurity and AI Board of Investigations"; H.R. 10484, the "Covered AI Prohibition Act"; H.R. 10502, a Director of National Intelligence report on AI use in section 702 queries; and H.R. 10515, a federal framework for AI in education and the workforce. These are statuses at capture, not final dispositions.
The same week showed that government AI deployment is itself contested: CNN, The Independent and The New Republic reported on the America.gov assistant, quoting it saying "Official results show Joseph R. Biden Jr. won the 2020 presidential election" and, after changes, "I do not provide political commentary." America.gov refused automated retrieval on September 30, so this thread is press record only; we have not observed the assistant.
Comuvia's judgement, labelled
This paragraph is a Comuvia judgement, not a finding from the record. We rate the complexity of this question high. Four mechanisms — voluntary industry practice, national encouragement with an auditor attached, standards co-written by the builders, and a development pause — are not variations on one policy; they imply different institutions, different enforcement and different people deciding. Collapsing them into "pro-AI versus anti-AI" loses what matters for a decision. We hold no position on which is correct, and nothing in this packet would support one. Not investment advice.
A note on form: the video companion uses AI-generated narration and a synthetic presenter, and says so on screen and in its description. The record fragments inside it are unaltered.