Nearly all speakers at the UN Security Council AI briefing, including Altman and Amodei, agreed on the policy measures Marcus has long called for.

Assessment: Unresolved Evidence as of 2026-10-28

Where and when Gary Marcus wrote it

Marcus on AI, Historic UN Security Council Briefing on AI, paragraph 10.

Said: 2026-09-23
Source observed: 2026-10-28live page, full text, public
Evidence as of: 2026-10-28the cutoff for this assessment
Assessment recorded: 2026-10-29assessment marcus-historic-un-security-council-briefing-2ddee6-a2

We recorded 8 claims from this post. See every claim from this post.

The wording above is the version this site observed on 2026-10-28. The source may have been edited after 2026-09-23. This page does not claim the wording was captured then.

What the claim commits to

Kind: Observation
Deadline, as the speaker wrote it: None stated. This site does not add one.
Deadline state when assessed: No horizon
Deadline state on 2026-10-29: No horizon (the release’s reference date)

From the speaker

Pretty much everyone
converged on exactly the same things
The need for preflight safety testing.
The need for transparency auditing.
The need for liability.
The need for global dialogue and international cooperation.
The need for immediate action.

Our interpretation

The claim describes substantive agreement at the September 23 briefing among nearly all speakers, including the four named briefers. It permits a small number of exceptions.
The five listed needs define the claimed convergence. Speakers need not use identical words, enumerate every measure, or endorse identical implementation details. Clear general assent to another speaker's proposals can count.
Preflight testing means safety assessment before training or deployment. Transparency auditing means disclosure and scrutiny sufficient to assess safety. Liability means legal responsibility for harms; general references to responsibility or human control do not automatically establish it.
Silence about a measure establishes neither agreement nor disagreement. Opposition to a global regulator does not itself oppose international dialogue or cooperation.
The phrase about what Marcus keeps advocating identifies the five measures. This assessment tests agreement at the briefing, rather than the history of his advocacy.

Counts as supported if: A sufficiently complete session record shows substantive convergence among nearly all speakers on the five listed needs, allowing a small number of exceptions and equivalent wording.

Counts as contradicted if: The session record establishes substantial disagreement on one or more listed needs, beyond the small number of exceptions allowed by 'pretty much everyone.' Alternatively, it affirmatively establishes that agreement was confined to a materially smaller group. One dissenter, institutional differences, or omitted topics alone do not suffice.

Retrospective: these criteria were written after the outcome was public.

Current assessment

Was it true for the period described?

Outcome: Unresolved

The record was checked, and the evidence available by the cutoff does not settle it.

Scored on a plain reading. The speaker named no measure for this claim, so it was judged on the reading an informed reader would take, stated below under our interpretation. Both reviews had to find that reading fair.

Evidence as of: 2026-10-28
Deadline state when assessed: No horizon
Assessment recorded: 2026-10-29
Policy version: 2026-10-28.7

Rationale

The records support important parts of Marcus's account: Bengio advocated liability insurance, Altman supported international cooperation, and Amodei proposed testing, outside scrutiny and urgent cooperation. They do not establish near-universal agreement on the entire five-part list, especially legal liability. Missing endorsements are not evidence of opposition. Nor does rejecting a global regulator establish rejection of international cooperation. The evidence therefore supports several shared priorities but does not settle the full claim.

Limitations

The main accessible record is an automatic transcript with transcription errors.
No numerical threshold was imposed on 'pretty much everyone.'
A broad shared-priorities reading is easier to support than agreement on every listed measure; Marcus does not fully specify that distinction.

Evidence (3 items)

Supports the claim: United Nations Transcripts, 2026-09-23
liability insurance
Why it matters here: Bengio explicitly endorses liability insurance. This supports one listed measure for one speaker, without establishing near-universal agreement.
Where
Yoshua Bengio's briefing, following his proposal to license frontier AI
Publisher
United Nations Transcripts
Kind of source
Secondary summary
Describes events of
2026-09-23
Public from
2026-09-23
Supports the claim: United Nations Transcripts, 2026-09-23
require cooperation.
Why it matters here: Altman supports cooperation, with surrounding proposals for safety standards and incident reporting. Elsewhere, the U.S. representative also supports dialogue while rejecting global governance; that institutional objection does not establish opposition to Marcus's cooperation measure.
Where
Sam Altman's briefing, discussion of democratic oversight and international standards
Publisher
United Nations Transcripts
Kind of source
Secondary summary
Describes events of
2026-09-23
Public from
2026-09-23
Supports the claim: The Singju Post, 2026-09-23
this urgent work.
Why it matters here: Amodei endorses urgent cooperation. His preceding proposals include external evaluators, common testing standards, verification and incident notification, supporting several of Marcus's themes without establishing agreement on liability.
Where
Briefing by Dario Amodei, concluding paragraph
Publisher
The Singju Post
Kind of source
Secondary summary
Describes events of
2026-09-23
Public from
2026-09-24

This claim is about OpenAI, and models from OpenAI helped review it: gpt-6-astra drafted it and gpt-6-astra reviewed it. Agreement between models is a check, not proof.

This claim is about Anthropic, and models from Anthropic helped review it: claude-opus-5-5 reviewed it and claude-opus-5-5 checked its final text. Agreement between models is a check, not proof.

AI models drafted and checked this assessment. gpt-6-astra (OpenAI) drafted it. claude-opus-5-5 (Anthropic) and gpt-6-astra (OpenAI) reviewed it independently, without seeing each other’s answers. claude-opus-5-5 (Anthropic) checked the final text against the frozen inputs. Code, not a model, decided whether it could be published.

Agreement between models is a check on the record, not proof that the assessment is right. Anthropic and OpenAI make these models, and some claims on this site are about those companies. How the method works.

What would settle it

An authenticated complete record or recording establishing speakers' substantive assent to the five needs, including liability, or showing substantial explicit disagreement beyond a small number of exceptions. A record that leaves those positions unstated may leave the claim unresolved.

Where it was said (1)

Said once in the sources this register covers.

2026-09-23: Marcus on AI, Historic UN Security Council Briefing on AI
Pretty much everyone converged on exactly the same things I keep harping on:
Where
paragraph 10
Published
2026-09-23
Source observed
2026-10-28, live page
Access
full text, public

History

Assessment versions

Assessment marcus-historic-un-security-council-briefing-2ddee6-a2: Current. Outcome: Unresolved. Evidence as of 2026-10-28. Public date not recorded.

Recorded events

2026-10-29
Assessment publishedBy this site
First assessment published after both reviews and the final-text check passed.
About assessment marcus-historic-un-security-council-briefing-2ddee6-a2.
2026-10-29
HeldBy this site
The openai review found the criteria unfair to the speaker's words.
About assessment marcus-historic-un-security-council-briefing-2ddee6-a1, which was not published.