Skip to content

Every Claim Checked Against Its Evidence · Human or AI

Stop misinformation
before it reaches anyone.

We build the guard between every claim and its audience. Written by a person or an AI, each claim is checked against the evidence it cites — and what the source doesn't support is stopped before delivery.

Try the Demo →
Supported — delivered with its evidence Unsupported — stopped at the guard Click the stream to send a claim through the guard

The Problem

Misinformation no longer looks like a lie. It looks like a footnote.1

Fluent.Confident.Sourced.1

It isn't new, and it isn't rare. In medical journals, 1 in 6 scientific citations misstates its source — and the rate hasn't improved in years.

Baethge & Jergas, Research Integrity and Peer Review, 2025 · 46 studies, about 32,000 citations

What One Word Did

  1. In 1980, a five-sentence letter to a medical journal reported that addiction was rare — in hospitalized patients.

  2. It was cited by 608 scientific articles.

  3. 4 in 5 of those citations dropped one word: hospitalized.

  4. That missing word helped fuel an epidemic. No AI was involved.

    Leung et al., N Engl J Med, 2017

Why Now

Now AI writes footnotes by the millions.
Trust has not kept up.

Elsevier, Researcher of the Future, 2025 · 3,200+ researchers in 113 countries

58%37% in 2024of researchers use AI in their work, up from 37% a year earlier
22%The Trust Gapconsider AI tools trustworthy
59%would trust AI tools more if they cited their references

Our Mission

End misinformation — whoever, or whatever, writes it.

What We Believe

  1. 01

    Every claim should carry its evidence.

  2. 02

    Confidence ≠ Proof.

  3. 03

    Show your evidence.

    The oldest standard in science, for people and AI alike.

Not who said it. What proves it.

Proof

Proven where it's hardest: science.

ProofSentry, our first product, checks scientific citations against the evidence behind them — whether a researcher wrote them or an AI did. Tested on 169,543 citation pairs from PubMed Central, 2020–2025.

96.5%

F1 Score With a Budget Model

Within 1.5 points of a frontier model handed the full text of every cited paper (98.0%).

4.3×

Fewer False Accusations

8.6 vs 36.8 per 1,000 correct citations, against GPT-5.5 xhigh with web search.

1.9×

More Complete Diagnoses

Right verdict, every error type, sound reasoning and traceable evidence — 75.0% vs 39.5%, against GPT-5.5 xhigh with web search.

Our Edge

Verification doesn't stop at the paywall.

When a cited paper is locked, ProofSentry convenes an Evidence Committee — the open papers that cite it — and checks the claim against what they report.

Where This Goes

Science is only the beginning.

One verification engine, delivered as an API that runs the moment before content ships. Publishers and research platforms first — then every product that delivers claims people rely on, written by people or by AI.

FromEvery ClaimPapers · AI Answers · Reports
ThroughThe GuardBetter Science Lab API
ToThe People Who Rely on ItReaders · Patients · Clients
Today Scientific Publishing Next Research AgentsMedicineLawFinanceNewsEnterprise Reports

The Trust Layer for Every Claim

In the age of AI, the scarcest resource is not information.

It's truth you can trace.