AI systems are making consequential decisions — loan approvals, medical triage, legal analysis. But there's no way to prove which model ran, what input it received, or what it actually produced. Verified AI anchors every inference to an immutable on-chain record.
AI outputs are currently unverifiable. Any bad actor can claim their output came from GPT-4o, Claude, or any other model. Verified AI closes that gap with cryptographic attestation anchored to Base.
AI systems make consequential decisions without any mechanism to verify which model ran, what input it received, or what it produced. AI accountability is theater.
Verified AI hashes both the prompt and response, generates a Groth16 zkML proof, and anchors everything permanently to Base mainnet in a single transaction.
Every attestation produces three immutable commitments stored in the on-chain registry.
// Three immutable commitments per inference
keccak256(prompt) → inputHash
keccak256(response) → outputHash
Groth16(π) → proofHash
// Stored in AttestationRegistry on Base
// Contract: 0x3dBF622ABC705d2Ec0E07EB0fCbb1AbFDe0281ebCryptographic commitment to the exact input. Cannot be retroactively changed or forged.
Immutable record of the exact output. Prove the model said exactly this — nothing more.
Groth16 zkML proof that the inference ran inside a verified TEE. Trustless, permissionless verification.
Add verified-ai-sdk to your project, or integrate directly via ethers.js / viem.
You need a Base mainnet wallet. Gas is ~$0.002 per attestation.
One-time call. Declare your TEE type and signing key in ProviderRegistry.
Hash model fingerprint + prompt + response. Submit the three commitments on-chain.
Permissionless read — no private key needed. Works from any environment.
// SDK — TypeScript / JavaScript
import { VerifiedAI } from 'verified-ai-sdk'
const client = new VerifiedAI({
network: 'base-mainnet',
privateKey: process.env.PRIVATE_KEY,
})
// Register your inference node (once)
await client.registerProvider({
name: 'My Inference Node',
teeType: 'AWS_NITRO',
publicKey: myTEESigningPublicKey,
})
// Attest an inference
const result = await client.attest({
model: 'gpt-4o',
prompt: userPrompt,
response: modelResponse,
})
console.log(result.id) // bytes32 on-chain ID
console.log(result.txHash) // basescan.org/tx/0x…All contracts are open source, MIT licensed, and verifiable on BaseScan.
Token is the key to the network. Stake to become a verified provider, vote on whitelist decisions, earn from attestation fees.
Community-first. Infrastructure follows.