Editorial standards

Our order of priorities

Accuracy comes before trustworthiness, trustworthiness before speed, speed before depth, and depth before presentation. When these conflict, we hold a story rather than publish something we cannot support.

Nothing is invented

Every factual statement in an article is tied to a numbered source. An automated quality gate checks each statement before publication and removes or blocks anything it cannot trace to the cited evidence.

• Figures, dates, product and model names must appear in the cited sources.
• Quotations must match the source word for word and are always attributed.
• Text copied from a source appears only inside quotation marks, and only briefly.
• We never invent sources, quotes, benchmarks, release dates, prices, capabilities or company statements, and we never create or alter images or videos.

Verification statuses

Each claim and each article carries a status. Only confirmed claims are presented as established fact; everything else is attributed.

• Confirmed — supported by primary evidence, such as the organization’s own announcement of its own action.
• Partially verified — some material claims are confirmed; others rely on limited or company-reported evidence.
• Developing — the story is new and evidence is still being gathered.
• Unverified — not independently confirmed. Shown with attribution, never as fact.
• Disputed — credible sources contradict each other on a material point.
• Corrected — a correction has been published.

Company claims are not independent evidence

When a company says its product is faster, better or more capable, we report that it said so. Such claims stay “company-reported” until independent testing supports them. News outlets repeating the same press release do not count as independent confirmation: our system groups republished copies together and counts them once.

Sources and their limits

We prefer primary sources: official announcements, documentation, release notes, repositories and regulatory publications. We also monitor independent reporting and technical analysis. A source’s reputation is a starting point, never proof — every claim is checked on its own.

Newsletters and tips can lead us to a story, but they are never treated as verification and their contents are never published.

Timestamps you can trust

Every article shows separately when the source published the information, when AI Newsroom first detected it, when verification finished, when the article was published, and when it was last updated. We never back-date detection or present a delayed discovery as real-time.

Human oversight

Before anything is written, an automated selection editor decides whether a development is both new and material: a new model, a new tool or capability people can use, a new technique or research insight, or a significant security incident. The focus is the technology, not the companies behind it: company news (funding, partnerships, policy) is published only when it is a major event. Routine changelog entries, enterprise administration features, documentation updates, minor releases, marketing and repeats are tracked but not published.

Depending on how this site is operated, selected stories are either approved by a human editor or published automatically once they pass every check. Stories with a failed quality check, disputed or weakly sourced claims, no primary source, wording the sources do not support, and all corrections always wait for a human editor. Each article states whether a human approved it.

Disclosure of automation

Articles are produced by an automated system. We disclose which engine produced each version — a language model (named, with its prompt version) or our template engine — and we never attribute AI-written text to a human journalist.

Media

We use official images and authorized embeds from the organizations involved, with attribution, and only after an editor approves them. We do not use unrelated stock imagery to decorate stories.

Untrusted content

Web pages we read are treated as data, never as instructions. Text inside a source cannot change how our system behaves or what it publishes.