VulcoraTalk to us
Security for the AI supply chain

Read what your AI is.
Not what it says.

The models, code, and agents entering your company look safe on the surface. We read what’s actually inside — and prove it.

Meet the instruments →Take the challenge →
or follow the dive, surface to abyss
Painted cross-section of the sea: warm daylight above the waterline fading to a black abyss below, with a single ember lantern sinking into the deep.
fig. 01 the descent, in cross-section — surface light to abyss
The problem

A hacked model passes every test — and turns on the one input you never tried.

It behaves perfectly until the day it sees the secret phrase it was trained on. No amount of testing will find that phrase — you have to look inside. And it’s not just models: the code your AI writes and the agents acting for you hide the same way. The surface always looks clean. And when a downloaded model goes wrong inside your product, the apology is yours to write.

What we do

So we read the real thing.

1 · Read

We read the inside

The model itself — not its paperwork. The wrapper is all anyone else reads.

2 · Prove

We prove every catch

Proof you can re-run yourself — never “trust us.”

3 · Or we say so

We never guess

When we can’t prove it, we tell you straight. Silence means unproven — never safe.

The scanner, at work

Watch a read happen.

This is what our scanner does — reads a downloaded AI straight from its file, without ever running it, and tells you in plain words what it found. It plays twice: first a model with a hidden backdoor, then a clean one.

One house, one rule

It senses, cures, defends, holds — and proves it.

Senses what a model really isRestores removed safety and excises the change in the weights — reversible, witnessed, coverage-boundedDefends the code your AI writesHolds agents to their wordProves every reading — machine-checked, clone-checkableMade in Solora — the workshop the whole house is built and watched from
the instruments of the house, side by side →
The head-to-head

Same threat. Different object.

The threatWhat the giants seeWhat Vulcora sees
A model trained with a hidden backdoora valid, correctly-signed file — cleared to shipthe hidden behaviour it will fire — and, where the read allows, a reversible fix
AI-written code that quietly leaks your datapasses the scanner as “clean”the exploit actually firing, with a proof you replay
An agent that stepped out of boundsa log of everything it didwhether it kept the promise it made before it started

It isn't that they're behind. They're pointed at a different thing.

See the scoreboard — our scores next to theirs →Or the full scored record, every miss included →
Why it matters

Your builders want the open model. Security could only ever answer “we can't be sure.” Vulcora is how you say yes — with proof in hand.

The house

The instruments of the house. One rule: prove it, or say nothing.

Each answers its own question about what's entering your company. Step through the one you need.

A luminous form resolving under a single shaft of light in the deep — reading what a model truly is.is it safe?

Protora

the trust read on a finetuned model

enter →A lone heron at the waterline against a thin band of horizon light — the watcher that names what it sees.what is this?

Ardora

the public model atlas

enter →A peacock perched in the dark, its hundred eyes turned outward — Argus, watching the code.is the code sound?

Pavora

proof-carrying code security

enter →A forge glowing in a dark timber hall — where work is bound to the record and proven.did it keep its word?

Askora

the accountable agent ledger

enter →can you prove it?

Proofora

the machine-checked formal proof

enter →where is it made?

Solora

the OS⊗IDE keystone

enter →
The journal

Follow the work.

Short, plain writing on what's actually inside your AI — and what we're up to. New posts as we go.

An antique field-journal etching: a plumb line and lead weight hanging in dark water, come to rest just above the seabed in lantern light — a floor found by feel.For builders

A bet on the floor, before the reveal

We entered a hard public estimation challenge and staked a specific number in advance — a floor no honest method should beat — with the core of the argument machine-checked. A prediction we would rather file, dated, than explain after the fact.

Jul 26, 2026 4 min readAn antique field-journal etching: a wax-sealed letter held before a lantern on the dark seabed, faint hidden writing showing through the paper — the seal never broken.For builders

We didn't just catch the hidden backdoor — we read out what it was told to do

The hardest backdoors are built so no test can ever trigger them — off-the-shelf scanners see nothing. We read the tampered model's own file and recovered the exact secret instruction, without setting it off.

Jul 23, 2026 4 min readAn antique field-journal etching: a sealed wooden crate on the dark seabed, a small lantern beside it casting warm ember light — what's inside is unknown.For everyone

Do you actually know what's inside the AI you use?

Most companies run AI they didn't build and have never read. Here's why the label on the box isn't proof — and what to ask instead.

Jul 21, 2026 4 min read
Read the journal →
One honest answer

Everyone tests what your AI says. Vulcora reads what it is — across your models, your code, and your agents — and proves it, or tells you straight when it can’t.

Talk to us →Request a demo

have a model you don't trust? send us its fingerprint — a short code taken from the file — the name can lie; the fingerprint can't.

Go deeper

Still curious? It goes deeper.

You're at the surface. From here the dive goes one layer at a time, each more technical than the last — take the next step, and stop whenever it's enough.

  1. You are hereLayer 1 · The surfaceThe idea in plain words — this page.0 m
  2. Next step — start the descent ↓Layer 2 · How it worksTriggers, witnesses, and the three verdicts.−200 m
  3. Layer 3 · The instrumentsThe instruments of the house, in product depth.−1 000 m
  4. Layer 4 · The disciplineHow attacks get in, and how we keep score in public.−4 000 m
  5. Layer 5 · The protocolsHow we lock an answer, prove a file, and package proof.−10 911 m

The deeper layers keep until you're ready — or see the whole dive on one sheet →