The AI News

Sources

Where the briefing reads.

The AI News reads a curated registry of 94 sources, organized into five source types below — the complete list, not a selection. New items usually enter the next hourly scan after a source publishes. Being monitored never guarantees publication: every story faces the same two-reviewer gate.

The registry prioritizes sources whose reporting can ordinarily be read without a subscription — the briefing should never send you to a locked door — and every published story links prominently to its original.

Primary research and official releases20

Labs, research institutions, agencies, and preprint servers — where announcements and results originate. Primary tellings are preferred when the same news arrives from several places.

Practitioner and expert analysis34

Independent researchers, builders, clinicians, and analysts. Practitioners add hands-on evidence and interpretation that announcements often lack; reputation earns attention, never automatic publication.

Engineering and applied research16

The teams building frameworks, infrastructure, and security tooling — where how-to-build knowledge and applied findings come from.

Independent reporting and trade press15

Journalism and trade coverage of the industry — lawsuits, chips, funding, policy, and incidents the primary sources never write up themselves.

Community and digest discovery9

Forums and curated digests where signal often surfaces first. Digest items restating already-covered stories are rejected by design.

How sources are chosen

A source earns its place by publishing original information or demonstrated specialist value, with reliable publication metadata, consistent AI relevance, ordinarily accessible articles, and enough substance per item to evaluate. Sources are removed for paywalls, repeated press-release rewriting, unreliable feeds or timestamps, persistently low signal, or inaccessible originals — the registry evolves, and no source pays for placement.

Reviewers also receive source-specific calibration notes. A few real examples: preprint servers carry “unreviewed — conservative significance until replicated”; vendor blogs carry “discount product marketing, keep genuine architecture lessons”; advocacy feeds are marked so their framing is read as perspective, not neutral reporting; and digests are told that restating covered stories fails novelty. Where coverage overlaps, the primary telling is preferred; original reporting and practitioner testing add evidence; secondary coverage passes only when it is the best accessible account.