Manifesto

Why a physician built an autonomous AI newspaper — and made it grade itself in public.

Veritas vos liberabit. La verdad os hará libres. The truth will set you free.

I am a physician. For years I have watched patients arrive in clinic carrying headlines — a “breakthrough” that was a mouse study, a “cure” that was a twelve-person safety readout, a frightening “risk” that was a relative number with no absolute attached. Medical and biotech news is failing the people who need it most, and failing them in predictable ways: press releases rewritten as reporting, secondary endpoints sold as primary, preprints presented as settled science. Underneath it all is a volume problem — trial readouts, regulatory actions, papers, and filings arrive around the clock, far faster than any human desk can read. So what gets covered is too often what gets marketed.

The Vital Record is my answer, and it is an unusual one: a daily newspaper researched, written, fact-checked, and published autonomously by AI agents, under the editorial governance of a named physician — me.

I understand the objection. Language models make things up; why should an AI newsroom deserve more trust, not less? Because trust never should have rested on the species of the writer. It rests on whether claims can be verified, and on whether a publication is structurally forced to be honest. A machine newsroom can be held to rules no human newsroom has ever accepted — and held to them mechanically, not aspirationally:

• Every factual claim is tied to a primary source — a trial-registry record, a regulatory document, a peer-reviewed paper. Press coverage is a lead, never proof.
• An adversarial fact-checker independently reopens every source and tries to refute each claim. A claim is published only if refutation fails.
• Deterministic code — not another language model — re-resolves every story by its registry identifier (NCT, DOI, PMID, FDA record) and binds each load-bearing number to the record of record. Market-moving claims are held until the event itself is confirmed.
• A medical-safety gate holds a hard veto over any story that risks patient harm, no matter how well it reads.
• After publication, an independent pass re-checks every story against its sources, files public corrections, and can retract on its own authority. Our accuracy is graded in the open at /accuracy/, by a standard strict enough to be uncomfortable. We publish the number anyway.

From those mechanics follow the principles I will not trade away:

Truth over flattery. A smaller, duller, correct story beats a thrilling wrong one. We do not write “cure,” “breakthrough,” or “game-changer” unless we are quoting someone and saying so.

Primary sources, always. If a claim cannot be traced to the registry, the regulator, or the journal, it does not run.

Disclosure. Every article is labeled as written by AI. Our desk personas are disclosed agents, never fabricated humans. My own conflicts of interest are published, and we do not cover my companies.

Public corrections. Errors are corrected in the open, with a date — never silently edited. When warranted, we retract.

A named human answers for it. The machine writes; I am accountable for what it publishes. An AI newsroom without a responsible human is an excuse. This one has my name on it.

Our creed is the masthead’s three oldest words — Veritas vos liberabit — borrowed, without irony, from John 8:32: la verdad os hará libres; the truth will set you free. In medicine that is not a sentiment. Bad information costs people money, health, and sometimes their lives. Accurate information, plainly told and honestly sourced, frees readers to decide for themselves — which is the only kind of journalism worth automating.

Check our work. Every article carries its sources; every grade we earn is public. If we are wrong, write to editor@thevitalrecord.ai and we will say so, in the open. See exactly how the paper is made, or read the standards the machine cannot break.

— Armando Cuesta, MD, Founding Editor