Edition 000001 · The Founding EditionSeptember 3, 2026Constitution 1.0

Editorial Standard

Source & Verification Standard

How sources are classified, material claims are verified, developer assertions are treated, and uncertainty is preserved.

Version: 1.0
Adopted: September 3, 2026
Publication: The Emergence Record


I. Purpose

The Emergence Record exists to preserve a trustworthy historical record of the development of artificial intelligence.

A factual claim is useful to the future only if readers can understand why the publication believed it was true.

For that reason, sourcing and verification are part of the permanent structure of the Record rather than an afterthought attached to publication.


II. Source Classes

Sources are classified according to their relationship to the information being reported.

Primary Sources

Primary sources originate directly from the person, organization, system, dataset, filing, paper, codebase, or institution responsible for the event or claim.

Examples include:

  • official company announcements;
  • research papers;
  • government filings;
  • court records;
  • legislation;
  • regulatory documents;
  • official repositories;
  • model cards;
  • technical reports;
  • financial filings;
  • direct transcripts;
  • published datasets.

Primary sources are preferred for establishing what an organization or individual officially did, released, claimed, or reported.

A primary source is not automatically correct about the broader significance or consequences of its own claims.

Independent Secondary Sources

Independent secondary sources provide reporting, analysis, or verification separate from the subject of the story.

Examples include reputable news organizations, independent researchers, academic institutions, and specialist publications with identifiable editorial accountability.

Supporting Sources

Supporting sources may provide useful context but ordinarily should not independently establish a major factual claim.

These may include:

  • industry blogs;
  • newsletters;
  • databases;
  • conference summaries;
  • social-media posts;
  • community discussions;
  • interviews reproduced by third parties.

Unverified Sources

Anonymous posts, unattributed claims, unsourced AI-generated material, screenshots without provenance, rumors, and inaccessible assertions are treated as unverified until corroborated.


III. Independence

Ten publications repeating one original report do not constitute ten independent confirmations.

The newsroom must identify the origin of a claim whenever possible.

When multiple sources ultimately rely upon the same original source, they should be treated as one evidentiary chain rather than separate confirmations.


IV. Verification Levels

Material factual claims may receive one of the following verification states.

UNVERIFIED

The claim has been identified but adequate supporting evidence has not yet been established.

It should not ordinarily appear as established fact in published reporting.

SUPPORTED

Credible evidence supports the claim, but independent confirmation may be incomplete.

CONFIRMED

The claim is supported by evidence judged sufficient for publication as established fact.

For major claims, this will ordinarily require either:

  1. a strong primary source plus meaningful independent confirmation; or
  2. multiple genuinely independent, credible sources.

DISPUTED

Credible sources materially disagree about the claim.

The disagreement itself should be reported.

FALSE

Available evidence demonstrates that the claim, as stated, is incorrect.

UNCERTAIN

Available evidence is insufficient to resolve the claim with reasonable confidence.

Uncertainty must be preserved rather than converted into false certainty.


V. Material Claims

Not every sentence requires the same level of verification.

A material claim is one that meaningfully affects a reader’s understanding of the event.

Examples include:

  • acquisition prices;
  • model capabilities;
  • benchmark results;
  • safety incidents;
  • legal decisions;
  • deployment claims;
  • dates;
  • technical breakthroughs;
  • casualties or harms;
  • financial figures;
  • statements attributed to people or organizations;
  • claims of firsts, records, or unprecedented capability.

Material claims should be stored individually in the Record and linked to the sources supporting or contradicting them whenever technically practical.


VI. Major Story Standard

Before a major news story is published, the newsroom should ordinarily establish:

  • the underlying event actually occurred;
  • material dates and figures;
  • the identities of relevant organizations or systems;
  • the origin of important claims;
  • at least one primary source when reasonably available;
  • meaningful independent confirmation when reasonably available;
  • whether significant contradictory evidence exists.

If these conditions cannot be met, the article must clearly describe the uncertainty.


VII. Claims Made by AI Developers

Claims made by organizations about their own artificial-intelligence systems require particular care.

Statements such as:

  • “state of the art”;
  • “human level”;
  • “superhuman”;
  • “safe”;
  • “aligned”;
  • “autonomous”;
  • “first”;
  • “breakthrough”;
  • “reasoning”;
  • “general intelligence”

must not automatically be adopted as factual descriptions merely because they appear in an official announcement.

The Record should distinguish between:

the organization says the system achieved X

and:

independent evidence establishes that the system achieved X.

VIII. Benchmarks

Benchmark results must be reported with sufficient context to prevent misleading comparisons.

When relevant, the Record should preserve:

  • benchmark name;
  • benchmark version;
  • score;
  • evaluation methodology;
  • prompting conditions;
  • tool availability;
  • sampling configuration;
  • comparison systems;
  • whether results were independently reproduced;
  • known contamination concerns.

A benchmark score is evidence about performance under specified conditions, not a universal measurement of intelligence.


IX. Social Media

Social-media posts may be primary sources for determining what an account publicly stated.

They are not automatically reliable evidence that the statement itself is true.

Screenshots should not be relied upon when the original post can be inspected directly.

Deleted or altered posts may be archived when their authenticity can be reasonably established.


X. Anonymous Sources

The Emergence Record should rely on anonymous sources sparingly.

When anonymous sourcing is necessary, the article should explain, to the extent possible without revealing the source:

  • why anonymity was granted;
  • how the source is positioned to know the information;
  • whether the information was independently corroborated.

AI systems must never invent anonymous sources, confidential conversations, interviews, quotations, or firsthand observations.


XI. Quotations

A quotation must be traceable to an identifiable source.

The newsroom will not fabricate quotations for narrative effect.

Paraphrased statements must not be presented inside quotation marks.

When transcript quality or translation creates uncertainty, that uncertainty should be disclosed.


XII. Automated Research

AI research systems may discover and analyze sources, but automation does not reduce the evidentiary standard.

An AI-generated summary of a source is not itself an independent source.

Whenever practical, claims should be verified against the underlying material rather than another model’s description of that material.


XIII. Conflicts of Interest

The model provider, infrastructure provider, database provider, hosting provider, sponsor, or other technology used to operate The Emergence Record receives no reduced verification standard.

If OpenAI technology contributes to an article about OpenAI, the provenance should remain visible.

The same principle applies to every provider.

Technical dependence does not create editorial allegiance.


XIV. Source Preservation

Whenever legally and technically practical, the Record should preserve enough metadata to allow a future reader to identify the source originally consulted.

This may include:

  • URL;
  • title;
  • publisher;
  • author;
  • publication date;
  • access date;
  • source classification;
  • archived location where appropriate.

The disappearance of a webpage should not erase the evidentiary history of the article.


XV. Corrections Resulting From Verification

If later evidence materially changes the truth status of a published claim, the Record should:

  1. preserve the original article version;
  2. update the claim’s verification status;
  3. publish a visible correction or annotation when material;
  4. identify the evidence responsible for the change;
  5. preserve the chronology of the correction.

Verification continues after publication when history demands it.


XVI. Governing Principle

The purpose of verification is not to create the appearance of certainty.

It is to preserve the strongest distinction possible between:

  • what we know;
  • why we believe it;
  • what remains uncertain;
  • and what later evidence changed.

The future deserves the evidence, not merely our conclusion.