The AI Judiciary is an independent product studio and publication system. We use a transparent court metaphor — claims, evidence, right of reply, rulings — because it makes reasoning inspectable. It is an editorial device only: we are not a legal court, government body or regulator, and we hold no authority over anyone.
“No vendor is the judge in its own case.”
Shared evidence standard, seal, typography and navigation.
Sourced adjudication of public AI claims
A model-neutral evidence court: one contested claim, the strongest case for it, the right of reply, independent evidence and a transparent ruling.
File the claim. Test the evidence. Read the verdict.
These are not working tools yet. They are published here so the direction is legible, not to imply functionality that does not exist.
Task-specific, evidence-led model comparison
Comparison built around a defined job rather than a leaderboard: fixed task, published protocol, visible limitations.
Claim intake and evidence triage
A structured way for readers to submit an AI claim they have seen, and for it to be triaged against the same evidence standard.
Monitoring emerging consequential AI claims
Tracking claims that are spreading quickly and would matter if true, before they harden into common knowledge.
Reopening cases when material evidence changes
A formal route back into a published ruling when new primary evidence arrives, so records improve instead of ageing badly.
No lab is a client. Any system can win or lose a case.
Every claim is tied to primary papers, protocols or documentation a reader can open.
The criticised party's published position is presented before any ruling is made.
“Not established” is written as “not established”, never rounded to true or false.
Each record ends with what it changes for a real person making a real decision.
A vendor's own measurement is testimony. It is never the verdict.
Future products may live on separate sites or domains. Wherever they run, they inherit this standard, the parent seal and the same disclosure about what this project is and is not.
AI-generated analysis. Written by AI in conversation with a user. Not an official statement, position or publication of OpenAI or any other AI vendor.
An evidence court for contested claims about AI. 10 cases are on the record.
Enter The Rebuttler DocketPrefer the method first? How we judge.