# agentage.org

> Editorial standards of Agentage, an AI-native news agency in preparation: sourcing, verification, AI labelling, human editorial responsibility and corrections.

Agentage is an AI-native news agency in preparation. agentage.org publishes its editorial standards.

## Principles

1. Facts before speed: We publish when a claim is verified, not when it is first seen. Being second and right beats being first and wrong.
2. Every claim has a source: Each factual statement links to where it comes from: a document, a dataset, an official record or a named publication. Readers can check it themselves.
3. Independent sources, counted: A news item rests on independent sources. When only one source exists, we say so in the text.
4. Our own words: We summarize and synthesize in our own wording. We do not copy other publishers' text; we link to the original and respect press publishers' rights.
5. AI is a tool, people are responsible: AI systems help us collect, compare and draft. A named human editor reviews and takes editorial responsibility for every published item.
6. Labelled AI involvement: Each item shows how AI was used: not at all, as research and drafting help, or as the main author under human review.
7. Corrections in the open: Errors are corrected on the same page with a dated correction note. We do not silently edit published facts.
8. No paid placement: Coverage cannot be bought. Sponsored content, if ever offered, will be marked as such and kept apart from news.
9. Separation of interests: The news operation is separate from the founder's consulting and training work. Conflicts of interest are disclosed in the item concerned.
10. Fairness to the people named: People and organizations named in a critical story get the chance to respond before publication, and their response is reported.

## AI labels

- A0 Human-written: Written and edited by people. AI tools were not used for the text.
- A1 AI-assisted: People wrote the item. AI helped with search, transcription, translation or comparison of sources.
- A2 AI-drafted, human-reviewed: An AI system drafted the text from verified sources. A named editor checked every claim and approved publication.

## Workflow

- Monitor: Public sources: official records, registries, press releases, data feeds.
- Collect: Store the original document and its address with a timestamp.
- Verify: Check the claim against independent sources; flag single-source items.
- Synthesize: Write in our own words; link every fact to its source.
- Review: A named human editor approves, labels AI use and signs off.
- Publish & correct: Publish with sources; corrections stay visible on the page.

## Questions

### What is Agentage?

Agentage is an AI-native news agency in preparation, founded by Zoltán Varga. It monitors public information, verifies it against independent sources and publishes synthesized reports in its own words. This site, agentage.org, publishes the editorial standards that every Agentage item follows.

### How does Agentage use AI in journalism?

AI systems help monitor sources, collect documents, compare claims and draft text. A named human editor reviews every item, checks the claims and takes editorial responsibility. Each item carries a label (A0, A1 or A2) that tells readers how AI was involved.

### Does AI-generated news need to be labelled in the EU?

Under Article 50(4) of the EU AI Act, AI-generated or manipulated text published to inform the public on matters of public interest must be disclosed, unless it has undergone human review or editorial control and a person holds editorial responsibility. Agentage labels AI involvement on every item and keeps human editorial responsibility in all cases.

### How are errors corrected?

Corrections are made on the same page with a dated note that explains what changed. Published facts are never silently edited. Readers can report errors through the contact page.

### Is Agentage independent from the founder's consulting work?

Yes. The news operation runs separately from Zoltán Varga's consulting and training work. Coverage cannot be bought, and any conflict of interest is disclosed in the item concerned.

### When does Agentage launch?

Agentage is in preparation. Launch dates and the first editions will be announced on the Agentage sites first: agentage.hu (Hungarian), agentage.news (English) and agentage.press (press releases).


## Founder

- Zoltán Varga — AI ecosystem architect and AI visibility researcher, Budapest: https://vargazoltan.ai/
- ORCID: https://orcid.org/0009-0003-1020-834X
- arXiv: https://arxiv.org/a/varga_z_1.html

## Articles

- [A0, A1, A2: How AI Involvement Labels on News Work](https://agentage.org/articles/ai-involvement-labels-a0-a1-a2/): News items should disclose AI involvement using a three-tier system: A0 for human-written content, A1 for AI-assisted research or transcription, and A2 for AI-drafted text with human review. This approach satisfies EU transparency requirements by focusing on editorial responsibility rather than relying on unreliable software detectors. Labels must remain accurate even after edits to ensure the origin of the information is always verifiable.
- [EU AI Act Article 50(4): The Human Review Exemption Explained](https://agentage.org/articles/eu-ai-act-article-50-4-news/): Under Article 50(4) of the EU AI Act, AI-generated text about matters of public interest must be labeled unless it undergoes human review with editorial responsibility. If a responsible editor reviews and approves the content before publication, the mandatory disclosure requirement does not apply. Without this human oversight, the content must carry a clear label indicating its artificial origin.
- [Press Publishers' Rights (DSM Article 15) for AI Newsrooms](https://agentage.org/articles/press-publishers-rights-dsm-article-15/): Under EU law, specifically the DSM Directive Article 15, an AI news service may freely reuse factual information and hyperlinks but cannot reproduce or rewrite the specific expression of a publisher's article without a license. While very short extracts and individual words remain free to use, any commercial reuse of the full text or substantial parts of a news article infringes the new neighbouring right held by the publisher.
- [How Visible Corrections Work: A Policy Guide for AI-Native News](https://agentage.org/articles/visible-corrections-policy/): News sites must replace silent edits with dated correction notes attached to the original article, explicitly stating what changed and why. When AI introduces errors like fabricated citations, the correction must trace the specific hallucination back through the audit trail to show the source of the mistake. This transparency preserves reader trust and meets emerging disclosure standards.
- [Crediting Another Outlet's Exclusive: A Checklist for News Agencies](https://agentage.org/articles/crediting-another-outlets-exclusive/): To credit another outlet's exclusive, you must explicitly name the original source in the text and clearly distinguish their specific claims from your independent verification. You should provide a hyperlink to the original report rather than reproducing their text, ensuring you add new context or data to avoid copyright infringement.
- [Why Labeling Beats AI Detection for News Trust](https://agentage.org/articles/labelling-versus-ai-detection/): No, current AI detectors cannot reliably determine if a news article was written by AI, especially when human editors have revised the text. Studies show detection accuracy drops significantly on hybrid content, rendering reader-side verification ineffective. The solution lies in publisher-side disclosure and provenance metadata rather than attempting to detect the output.
- [Respecting Text and Data Mining Opt-Outs: Collection Ethics](https://agentage.org/articles/text-and-data-mining-opt-outs-collection-ethics/): An AI news agency may crawl and analyze other publishers' sites only if the publisher has not issued a machine-readable opt-out via robots.txt or a specific TDM signal. Under the EU Text and Data Mining exception, agencies can store full texts strictly for internal analysis, but they must never republish the original content or substantial parts of it without explicit licensing.
- [Reporting from Paywalled Sources: What Is Allowed Under EU Law](https://agentage.org/articles/reporting-from-paywalled-sources/): A news agency may report the facts contained within a paywalled article, such as who did what and when, because facts are not protected by copyright. However, the agency cannot summarize the specific text, arguments, or unique phrasing found behind the paywall without explicit permission or a license. You must attribute the discovery of the fact to the original source while writing the summary entirely in your own words.
- [Editorial Standards Glossary for AI-Native Journalism](https://agentage.org/articles/editorial-standards-glossary/): In AI-native journalism, editorial responsibility means a human editor must verify and approve content before publication to satisfy EU law. A single-source item requires cross-verification with independent data to prevent misinformation, while provenance tracks the origin of every fact and image used. These standards ensure that automated systems serve as tools for synthesis rather than replacements for human judgment.
- [AI Journalism Standards: Liability, Labels, and Editorial Ethics](https://agentage.org/articles/ai-journalism-standards-faq/): An AI-assisted news agency must maintain human editorial oversight to assume legal liability, clearly label all AI-generated content with visible disclosures and machine-readable metadata, and establish a public correction protocol that updates labels if the content is significantly altered post-publication. These standards align with the EU AI Act transparency requirements and financial compliance rules as of April 2026.
- [Signing What We Publish: Content Provenance as Policy](https://agentage.org/articles/content-provenance-signing-published-news/): News organizations prove origin by embedding cryptographically signed metadata into files using standards like C2PA, which records the creator, device, and edit history. This creates a tamper-evident trail that allows readers and platforms to verify if content was altered after publication. Legal frameworks like the EU AI Act now mandate such transparency for generative AI outputs.