The reasoning
attests itself.
AI is starting to act on the physical world — pricing carbon, grading gold, moving capital — and no one is held to what it claims. VPAY makes an AI verdict accountable: the model commits its output before the result is known, posts a slashable bond, and anyone can challenge it and recompute. Not a promise the model is right — a stake that it can be checked.
Adwene di adanseɛ — the mind bears witness.
Autonomous AI, acting on the world,
accountable to no one.
The hard problem isn't whether a model is correct — it's that when an agent makes a claim about a real asset, there is no record of who claimed what, no stake behind it, and no way for an outsider to challenge it. Accountability, not correctness, is the missing layer.
No authorship
An agent issues a verdict and the trail vanishes. Which model, which inputs, which moment — unrecorded.
No stake
A wrong claim costs the model nothing. Without a bond, there is nothing to lose and nothing to slash.
No challenge
Outputs arrive as "trust me." No outsider can independently replay and dispute the claim.
Commit first. Reveal after. Recompute always.
Before it sees the outcome, the model seals a commitment: which weights, which context, which seed, which output. It posts a bond. The window opens for anyone to challenge — and the seed reveal lets them replay it independently.
POLYGON · 137
7EA38E92<4F1CA930<BONDED<CHALLENGEABLE<<<<4
Illustrative of the live KommitBridge tuple. The contract is deployed and source-verified on Polygon Mainnet.
From a model's claim to a stake anyone can test.
Four moves. The commit and the bond are live on Polygon today — the same registry the physical proofs anchor to.
Seal before outcome
The model hashes its weights, context, seed and output and commits the tuple on-chain — before the result is known.
Post a stake
A slashable bond is posted behind the claim. Being wrong, once challenged, has a price.
Open the window
Anyone can post a counter-bond and dispute the claim during the challenge window.
Replay & resolve
The seed is revealed; the output is independently recomputed. The honest side keeps the bond.
The contract is live.
Routine attestation is next.
The AI metrics we publish are not "trust us" numbers. Each snapshot is hashed into a Merkle root and anchored to the public AnchorRegistry on Polygon — so anyone can recompute the hash and confirm the published figures are the anchored figures.
Builders of agents that act on real assets —
let's make them accountable.
If your AI makes claims that money or lives depend on, the commit → bond → challenge → recompute layer is built and live. Integrate it, or pressure-test it.