The record lives on the instruments.
There is no separate ledger restated here. The graded record is kept where it is made — these are the doors into it. Every line behind them can be re-run by a stranger; until a result carries its proof, it doesn't get a line.
In one breathEach door opens onto a record you can re-run yourself, kept on the instrument that made it.
The sealed challenge
We published several open AI models — one with a hidden trap inside — and dared the world to find it. Before it opened, we locked the answer in by publishing its fingerprint first, so nobody (us included) could quietly change it later. The usual scanners spot none of the traps.
See the challenge →The catalogue of catches
A public list of the backdoored models we've caught — each with proof you can re-run yourself, shown next to how the usual scanners scored it. They miss every one. And where we've missed something, that's on the card too.
Browse the catalogue →The public exams
We grade ourselves on the same public tests researchers use, and against the free scanners anyone can download. The posted scores — PADBench, NIST TrojAI, and BAIT-ModelZoo — carry proof, failures included; BackdoorLLM and TDC 2023 are surveyed but have no honest offline path, and the record says exactly that.
Check the scores yourself →The continuous read
We're always reading the open AI models people actually download and use, and mapping them in a public library. When we find something real, the model's owner hears it first, in private. The record grows here.
The model library →the owner hears it first, always · we read the whole field, not one fileThe attestation plane
One URL, and it answers a stranger in JSON with no account and no permission: Does Vulcora hold a signed attestation dossier for this model? Either the signed dossier itself, or one of 8 named reasons we serve none — the vocabulary is closed so a refusal can be audited rather than shrugged at, and 3 of the 8 concede that we hold signed evidence we will not serve: it disagrees with what we publish, it disagrees with itself, or it carries no verdict inside the signature at all. Its index publishes the gap between the models we render a verdict about and the models we hold signed evidence for, and hands you the one-line command that computes that gap on your own machine.
Open the plane →the keys are served by the same host as the dossiers — the anchor is circular, so a passing check is integrity-onlyIn one sentence: we read what a model was trained to do — straight from its file — and prove it, or say we can't.