AI Newsroom Disclosure Profile

A standard for how an automated newsroom discloses how its material was produced. Written to be adopted by newsrooms whose writers are people, whose writers are models, and whose writers are both.

ai-newsroom-disclosure-profile/1.0 версія 1.0 4 обов'язкових пунктів 4 заборон 5 артефактів

Область дії

any newsroom that publishes material generated by a model

This profile governs disclosure, not verification. It says what a reader must be able to see, not what a reader must believe.

Профіль не регулює:

editorial judgement about the truth of a claim
the reporter's choice of what to cover

Обов'язкові пункти

JSON →

what_was_done

State that the text was generated automatically from the sources shown, and name the agent that checked it.

article_page app/disclosure.py:ARTICLE_LABEL

who_is_responsible

Name the accountable party. Where no human editor exists, name the process and make it auditable.

article_page app/disclosure.py:ACCOUNTABILITY

what_is_not_confirmed

Show the uncertainty the newsroom recorded, rather than hiding it.

article_page app/disclosure.py:build_disclosure

where_to_check

Give the reader the evidence: the sources above the text, the verifier's notes below it, and a public corrections log.

article_page app/disclosure.py:EVIDENCE_LINKS

Заборони

no_bare_ai_label — Do not label content with a bare «AI-generated» or «AI-written» mark. Name the action, not the technology. підстава: altay-gilardi-2024
no_claim_of_human_review — Do not claim human review, human editing, or human oversight unless a person actually did it. підстава: ali-2025
no_verdict_without_evidence — Do not publish a verification verdict without the evidence that supports it. підстава: schilke-reimann-2025
no_silent_edits — Do not change published text without recording the change in a public corrections log. підстава: toff-simon-2025

Що має публікувати редакція

Обов'язкові артефакти стандарту
Артефакт	Вимога	Де	Реалізація
sources	The source list for the piece.	`article_page`	`article.sources`
verifier_notes	The checking agent's findings, including its objections.	`article_page`	`article.verifier.issues`
provenance	A signature over the published text, so a third party can confirm it was not changed after verification.	`provenance_page`	`app/provenance.py:manifest_for_article`
corrections_log	A public log of every change made after publication.	`corrections_page`	`app/db.py:list_corrections`
decision_journal	A public record of the decisions taken in place of a person, so the absence of a person is auditable rather than decorative.	`arbiter_page`	`app/uarp_arbiter.py`

Дослідження в основі

Дослідження →

Altay & Gilardi, PNAS Nexus, 2024 N=4976 — A generic «AI-generated» label lowers perceived accuracy and sharing intent even when the content is true.
Toff & Simon, IJPP, 2025 — Disclosing the source list largely cancels the trust penalty that an AI label creates.
Ali, 2025 N=580 — A vague label costs about 0.55 trust points; a process disclosure removes the penalty, and adding an accountability clause beats no disclosure at all.
Schilke & Reimann, OBHDP, 2025 — Disclosure erodes trust unless it is paired with responsibility and monitoring cues.

Джерела — у дослідженні, де вони розібрано докладніше разом з аудитом нашого функціоналу.

Відповідність

conformant — All required clauses, artifacts and prohibitions are met, and the profile version is stated.
partial — The required clauses are met, but one or more artifacts are not published. The gap must be named.
non-conformant — A required clause is missing, or a prohibited claim is made.

A newsroom declares conformance by publishing the profile identifier and version it complies with, next to the material it publishes.

Приклад декларації: Disclosure: ai-newsroom-disclosure-profile/1.0 (v1.0)

Специфікація як JSON

Це не текст для читання, а специфікація для виконання. Інша редакція може завантажити її, перевірити свою практику на відповідність і заявити про відповідність із номером версії.

/api/disclosure-profile · /api/manifest · живий маніфест →