Every AI output.
Cryptographically proven.
Signet generates a 152-byte hardware-signed receipt for every AI inference — binding exact model, output, and timestamp in a tamper-evident record.
Verifiable offline forever. No vendor. No trust required. One receipt proves everything.
AI deals get blocked because auditors want proof.
No audit trail
AI vendors give you an output. Nothing proves which model version signed it.
Compliance blocks
EU AI Act, SEC 17a-4, FDA PCCP all require traceable AI output logs. Today you can't produce them.
Vendor lock-in
Every cloud's attestation is proprietary. Switch providers and your audit trail is gone.
Three steps. One receipt. Forever verifiable.
Generate
Call aetherproof.sign() after any inference. One line of code.
Seal
152-byte receipt signed with Ed25519+ML-DSA-65, hardware-rooted at TPM2 or NVIDIA CC.
Verify
Anyone verifies offline forever. No Signet servers. No vendor SDK. Just the receipt and public key.
Watch a receipt get generated.
Receipt Preview showing real receipt output format. Run the CLI for production use.
$ aetherproof generate --model gpt-4o▌
What nobody else does.
| Tool | What it proves | Per-output | Hardware-rooted | Offline verifiable | Cross-cloud |
|---|---|---|---|---|---|
| Vanta / Drata | Policy compliance | ○ | ○ | ○ | ○ |
| Global Relay / Smarsh | Comms archive | ○ | ○ | ○ | ○ |
| EQTY Lab | Build/load-time env | ○ | ● | ○ | ○ |
| AWS Nitro Enclave | Enclave image hash | ○ | ● | ○ | AWS only |
| Sigstore / Rekor | Code signing | code only | ○ | ● | ● |
| SignetCryptographic proof for every inference output and the model that produced it — verifiable offline across any cloud. | Per-inference output + model identity | ● | ● | ● | yes (normalizer) |
Already mapped to every major framework.
The protocol, in motion.
Get in before the queue closes.
Signet is in closed early access. We're onboarding compliance teams, AI vendors, and infrastructure builders first.