Hive
Hive Proof Architecture
The obligation is written down. The portable instrument that discharges it is still an open surface, and that is a question of timing.
Stripe has built the control layer for agentic commerce: scoped tokens, live limit enforcement, Radar decisioning, agent identity, and approval rules with a human in the loop. Hive is an additive, removable proof sidecar that binds an independently signed artifact beside those events, so a third party can check later, offline, that the authority existed, what its exact scope was, and that the action fell inside it. Stripe's route, authorization, limits, decisioning and merchant of record all stay exactly as they are.
Both the Agentic Commerce Agent Services and Seller Services terms are marked Preview and were last modified April 20, 2026 (stripe.com/legal/ssa-services-terms).
“retain auditable records of User's Authority at the time of each Agentic Transaction.”
“the Agentic Transaction was not authorized by the Agentic Customer, exceeded User's Authority, or was caused by bugs, hallucinations or misinterpretations” and, separately, a transaction generated or initiated “due to error, malfunction, unauthorized action, or the agent acting without requisite authority or outside the scope of its permitted authority.”
“Stripe may develop a framework for identifying trusted agents. To qualify as a trusted agent, User must comply with the criteria for trusted agents that Stripe provides to User or publishes.”
Read through, in one line: the duty to prove the customer's authority at transaction time is already in writing, and the portable artifact that discharges that duty in front of a counterparty, an auditor or an insurer is the surface still open. Definitions in the same document set Authority as “the express scope of authority an Agentic Customer grants to any AI agent… including any applicable spending limits, merchant restrictions, or time-bound parameters” (Stripe terms). That definition is exactly the field set a signed authority record binds.
The right column is additive. Every item in the left column stays authoritative, stays in Stripe's hands, and keeps working exactly as it does today.
usage_limits with currency, max_amount and expires_at, scoped to one seller's Stripe profile (Stripe docs).“But even if the app is ready to accept agentic payments, how do we give an agent a payment credential in the first place and how can we trust them to use it responsibly?”
“spend funds that it shouldn't authorize.”
“When there's an agent between you and the API, everything changes: how products are discovered, behavior is verified, and trust is enforced.”
“trust can't be inferred, it has to be explicitly granted, scoped, and enforced in code.”
Hive's position is one sentence longer than that last line: granted, scoped, enforced in code, and then provable to someone who was not there. Two neutral voices from the same Stripe stage put the same point in their own terms: Guillaume Poncin, CTO, Alchemy, on agents, “you actually need a verification moment. You need to close the loop”, and Manik Surtani, CTO of the AAIF, Linux Foundation, on the trust layer, “it's still nascent. It's still too young” (both Sessions 2026).
Stripe would own the customer facing proof product, its packaging, its price and the relationship. Authorization, limit enforcement, Radar decisioning, settlement and merchant of record stay authoritative and unchanged. Nothing moves out of Stripe's control. This is a proposed structure, not a current commercial arrangement.
The signer and verifier underneath, so the artifact a counterparty checks has cryptographic separation from the operating record it references. Independence is the product. A receipt Stripe signs about itself proves less in a dispute than one signed by a party with no economic stake in the outcome.
The agent action path runs across the top and stays Stripe's. Hive receipts bind underneath as each step passes. Switch Hive off and watch what happens to the path.
Hive is a sidecar. Hive fails open. If Hive is unavailable the payment proceeds and nothing is blocked. Take Hive out and Stripe's behavior is identical to today: there is no shim to unwind, no policy to restore, and no settlement dependency to migrate, because Hive never held any of them. Previously issued receipts remain independently verifiable offline from the artifact itself, against published key material, with no call to Hive.
Timing favours a native fit. The Agentic Commerce Protocol is at API version 2026-01-30, which added capability negotiation, a payment-handlers framework and an extensions framework (ACP changelog), and the specification is still marked draft (ACP repository). An evidence extension therefore arrives as a native extension rather than a retrofit. And because Shared Payment Tokens already extend to Mastercard Agent Pay, Visa Intelligent Commerce, Affirm and Klarna (Stripe blog), an evidence layer beside the token sits at network scale rather than at one processor.
Three states only, and nothing on this page is upgraded past them. LIVE means a Hive production endpoint answers today. BUILT AND TESTED, NOT DEPLOYED means the artifact exists with passing tests and no deployment. PROPOSED means designed for this brief and offered as proof of concept work.
Every matter below is a public court or regulator document, linked to the primary source. None of them involve Hive, Stripe, or any Hive customer. They are here for one reason. In each one the contested question was not whether the system worked. It was whether anyone outside the operator could check the record afterwards. The right hand column names the canon entry built for that exact failure. Maturity for each entry is stated in the instrument map above and in the canon explorer, and nothing on this page upgrades it.
Four honest notes. Cartisim is a complaint, so those are allegations that were never tested at trial. LeadClick was reversed as to the parent company, so it is cited only for the question of who authored the pages. Paddle and Nexway are consent documents in which the conduct was neither admitted nor denied. Robinhood and TSB are regulator findings about operational and systems records, not findings that a published incident report was wrong. Nothing above is a Hive claim, and none of it is evidence that Hive would have changed any of these outcomes. It is offered only as the public record of how often the contested fact turns out to be the record itself.
Pick a boundary. Stripe stays authoritative for the money event, authorization, limit enforcement, Radar decisioning and merchant of record throughout.
If a buyer cannot prove the claim without a meeting, the demo is what is broken, not the buyer's calendar. Everything below runs against production right now. Paste it into a terminal.
That is a signed authority record with an SPT shaped scope and a 250 dollar cap, checked by a verifier that needs no key and no account. You get valid: true and nine individually reported gates.
Nobody raises that limit after the fact, including us. If you would rather cut Hive out of the check entirely, the Ed25519 public key is published at /keys and you can validate the signature inside your own process with any standard library.
This list is published for the same reason the receipts are open. An evidence vendor that overstates its own maturity has disqualified itself from the category it is selling into.
There is also a published program path: the Agentic Commerce Protocol program lists [email protected] as its contact address (agenticcommerce.dev). One surface, one owner, one message is enough to start.
A suitable scope is one x402 on Base route for 30 days, with a signed receipt bound beside each machine payment. Stripe's route, price, limits, decisioning and settlement logic stay exactly as they are.
The integration scope is limited to the proof fields and the signing path Stripe chooses to expose, and Hive removes itself on request at any point.
These receipts add a precise record of screening and a repeatable check that a single agentic transaction fit one delegated mandate beside the authority, limits, divergence, parity, and sanctions themes already described here. They run in production today. The examples below are verified live against them.
This receipt records the exact screening context associated with an agentic payment or counterparty. It binds the named engine, ruleset, reference lists, list versions, screening time, validity period, and declared verdict. The subject identifier and list records remain protected as commitments, while risk and compliance teams can inspect the evidence path later. A matched outcome remains tied to the screening context that produced it. That gives counsel and risk teams a portable record beside the payment authority evidence.
What it does not do. It does not show that any list is complete, accurate, current, or free of omissions, decide that a counterparty is actually restricted, catch every false negative, catch an engine that clears a match it found, or assess the adequacy of the program, its thresholds, lists, or validity period. It does not identify the counterparty, expose a list record or match score, prove custody of the commitment key or window secret, or admit, block, freeze, reverse, settle, report, or provide a legal, regulatory, or compliance conclusion about any transaction.
Nothing has run yet. Click Run it and the answer below comes back from the verifier, not from this page.
Nothing has run yet. Click Run it and the answer below comes back from the verifier, not from this page.
Nothing has run yet. Click Run it and the answer below comes back from the verifier, not from this page.
This receipt compares a named agentic transaction with one delegated authority receipt signed before the transaction was authorized. It checks the declared amount, currency, timing, and scope against the constraints in that mandate. The service recomputes the result instead of accepting a caller supplied answer. That gives a reviewer a clear answer about whether this one transaction fits the supplied delegated limits. It can sit beside the existing authority and payment records without becoming part of the payment decision.
What it does not do. It does not prove that a cardholder granted the delegation, that the agent identity is genuine, that a network authorized or settled the transaction, that goods or services were delivered, or that a person read the displayed terms. It is not payment authorization, holds no cardholder credential, has no recognition as authentication data, compelling evidence, or liability shift from a network, issuer, or regulator, cannot resolve or affect a dispute or consumer rights, cannot assess aggregate spend or velocity, and cannot show that the supplied mandate remained unrevoked when the transaction was authorized.
Nothing has run yet. Click Run it and the answer below comes back from the verifier, not from this page.
Nothing has run yet. Click Run it and the answer below comes back from the verifier, not from this page.
Every run above posts a verified request body from this domain to the open verify route and prints what came back. The example receipts are signed with published example keys, so verify reports key_trust example_registry. That is on purpose. Nothing on this page is a production issuance, a customer record, or an endorsement. Patent Pending.
Each link opens that entry in the canon implementation explorer, where its schema, mint route, open verify route, auth requirement and implementation state are stated. The state shown here is read from the same registry file the explorer renders from, so the two cannot drift apart. Nothing here implies a customer, a deployment or an endorsement.