Why architecture matters here
AP2 chargebacks matter because disputes are inevitable in any payment system, and agent-initiated payments add a critical new question -- was the agent's payment properly authorized? -- that determines liability and requires the mandate as evidence. Any payment system has disputes (fraud, unauthorized charges, quality issues -- users disputing charges, chargebacks reversing them). Agent payments add a new dimension: the payment was made by an agent (on the user's behalf), so a dispute raises the question of whether the agent was properly authorized to make it (did the user actually authorize this payment, via a valid mandate within its scope?). This question (agent authorization) is central to agent-payment disputes (determining whether the payment was legitimate -- properly authorized by the user's mandate -- or not) and to liability (who bears the loss if it wasn't authorized). Handling this well (with the mandate as verifiable evidence of authorization) is essential to agent payments being trustworthy (users trusting that unauthorized agent payments can be disputed and reversed; merchants/operators trusting that authorized payments can be defended). For AP2 (agent payments) to be viable, robust chargeback/dispute handling (centered on the mandate as authorization evidence) is essential, and understanding it is understanding how agent-payment disputes are resolved.
The mandate-as-evidence insight is the crux, and it's what makes agent-payment authorization verifiable. In a dispute, the key question is whether the agent's payment was authorized by the user. The mandate answers this: it's the user's cryptographically-verifiable authorization that delegated the payment capability to the agent (specifying the scope -- what the agent can pay for, up to what amount, under what conditions). So the mandate is the evidence of authorization: if the disputed payment was within the mandate the user granted (the agent paid for something the mandate authorized, within its limits), the mandate is strong evidence the payment was authorized (the user cryptographically authorized this category of payment) -- so the payment is legitimate (defensible against the dispute). If the payment was outside the mandate (the agent exceeded its scope or limits, or there was no valid mandate), the payment wasn't properly authorized (the mandate doesn't cover it) -- so the dispute is valid (the payment should be reversed, and the liability falls on whoever allowed the unauthorized payment). This makes the mandate the linchpin of agent-payment disputes (the verifiable evidence of whether the payment was authorized) -- versus traditional payments (where authorization is murkier -- was it really the cardholder?). The cryptographic, scoped mandate provides clear, verifiable evidence of agent-payment authorization. Understanding the mandate as evidence (the verifiable authorization determining whether the agent's payment was legitimate) is understanding the crux of AP2 chargebacks.
And the agent-vs-human-liability dimension is the novel, crucial concern, because it introduces new liability questions. Traditional chargebacks have established liability rules (roughly -- the merchant, issuer, or cardholder bears the loss depending on the case and the authentication). Agent payments add a new dimension: the payment was made by an agent, so if it's unauthorized or erroneous, who is liable? Several new questions arise. Did the user's mandate cover the payment (if yes -- the user authorized it -- the user likely bears responsibility, like authorizing any payment; if no -- the agent exceeded its authority -- the liability may fall on the agent operator, whose agent made an unauthorized payment)? Did the agent malfunction (an agent error making an erroneous payment -- raising the agent operator's liability)? Was the mandate itself compromised (a security failure -- raising different liability)? So agent payments introduce new liability questions (agent operator liability for unauthorized/erroneous agent payments, versus user liability for properly-mandated payments) -- a dimension traditional chargebacks don't have (no agent intermediary). Resolving these (with the mandate determining whether the payment was within the user's authorization -- and thus whether the user or the agent operator bears liability) is a novel, crucial concern of agent-payment disputes. Understanding the agent-vs-human liability dimension (the new liability questions agent payments introduce, resolved via the mandate) is understanding a crucial novel aspect of AP2 chargebacks.
The architecture: every piece explained
Top row: the problem and the flow. The problem: an agent-initiated payment is disputed -- raising the question of whether the agent was properly authorized to make it. Chargeback flow: a dispute (the user or cardholder disputing the payment) leading to a potential reversal (the payment reversed if the dispute is upheld) -- the dispute-to-reversal process. Mandate as evidence: the user's cryptographically-verifiable authorization (the mandate -- delegating the payment capability to the agent, with scope/limits) as the key evidence of whether the payment was authorized. Liability: who bears the loss (the user, agent operator, merchant, or payment provider) -- determined by the authorization (the mandate) and the dispute reason.
Middle row: reasons, evidence, liability. Dispute reasons: the reasons for disputes -- fraud (fraudulent payment), not-authorized (the payment wasn't authorized -- e.g., the agent exceeded its mandate), quality/non-delivery (the goods/services weren't as expected) -- different reasons with different evidence/liability. Evidence trail: the evidence for resolving the dispute -- the user's intent (the original intent that led to the payment), the mandate (authorizing the agent), the transaction receipt (the payment record) -- the trail proving (or disproving) authorization. Agent vs human liability: the novel dimension -- if the payment was unauthorized/erroneous, who's liable (the user, if the mandate covered it; the agent operator, if the agent exceeded its authority or malfunctioned) -- determined by whether the payment was within the mandate. Representment: contesting the chargeback (the merchant/operator presenting evidence -- the mandate proving authorization -- to contest the dispute) -- defending a legitimate payment.
Bottom rows: prevention and comparison. Prevention: reducing disputes via verifiable mandates (clear, cryptographic authorization -- so payments are properly authorized and defensible, reducing not-authorized disputes) -- prevention through clear authorization. vs traditional chargebacks: AP2 chargebacks extend traditional chargebacks with the agent-authorization dimension (the mandate as evidence, agent-vs-human liability) -- the agent-specific issues (whereas traditional chargebacks lack the agent intermediary). The ops strip: evidence retention (retaining the evidence trail -- intents, mandates, receipts -- so disputes can be resolved with the authorization evidence -- crucial for representment), dispute handling (the process for handling disputes -- investigating, gathering evidence, representment or accepting the chargeback), and risk (managing the dispute/chargeback risk -- fraud prevention, verifiable mandates, monitoring dispute rates -- the risk of agent-payment disputes).
End-to-end flow
Trace a disputed agent payment. An agent makes a payment on a user's behalf (within a mandate the user granted -- authorizing the agent to pay for a category of goods up to a limit). Later, the user disputes the payment (a chargeback -- claiming they didn't authorize it). The dispute resolution centers on the mandate (the evidence of authorization): the merchant/operator presents the evidence trail -- the user's original intent, the mandate (cryptographically authorizing the agent to make this category of payment, within the limit -- and the payment was within it), and the receipt. Since the payment was within the user's mandate (the cryptographic authorization the user granted), the mandate is strong evidence the payment was authorized (the user did authorize this -- via the mandate) -- so the operator can contest the chargeback (representment -- presenting the mandate as proof of authorization). If the mandate clearly covers the payment, the dispute is likely resolved in the operator's favor (the payment was authorized -- the mandate proves it) -- versus a traditional payment where authorization is murkier. The mandate (verifiable authorization evidence) was the linchpin of resolving the dispute (proving the agent's payment was authorized).
The unauthorized and liability vignettes show the harder cases. An unauthorized case: an agent makes a payment OUTSIDE its mandate (exceeding the scope or limit the user granted -- e.g., paying for something the mandate didn't authorize, or above the limit). When disputed, the mandate does NOT cover the payment (the evidence shows the payment exceeded the user's authorization) -- so the payment wasn't properly authorized (the dispute is valid -- the payment should be reversed). And the liability question: since the agent exceeded its authority (made a payment the user's mandate didn't cover), the liability may fall on the agent operator (whose agent made the unauthorized payment -- not the user, who didn't authorize it) -- the agent-vs-human liability resolved by the mandate (the payment outside the mandate -> the user didn't authorize it -> the operator, whose agent overstepped, bears liability). A liability case: an agent malfunctions (a bug causing an erroneous payment). The mandate might technically cover it (within scope) but the payment was erroneous (the agent malfunctioned) -- raising the agent operator's liability (their agent's error) -- a new liability question (agent malfunction) that the operator must handle (bearing the liability for their agent's error, and fixing the bug).
The prevention and comparison vignettes complete it. A prevention case: the team reduces disputes through verifiable mandates (clear, cryptographic authorization -- so payments are properly authorized within well-defined mandates, and the authorization is verifiable) -- reducing not-authorized disputes (the payments clearly authorized and defensible) and making the ones that occur resolvable (the mandate as clear evidence). Clear mandates (verifiable authorization) prevent and ease disputes. A comparison case: the team notes that AP2 chargebacks extend traditional chargebacks with the agent dimension -- the mandate as verifiable authorization evidence (clearer than traditional authorization), and the agent-vs-human liability question (novel -- who's liable for an agent's unauthorized/erroneous payment) -- so their dispute handling accounts for the agent-specific issues (the mandate evidence, the agent liability) beyond traditional chargeback handling. The consolidated discipline the team documents: center agent-payment disputes on the mandate (the cryptographically-verifiable authorization -- the key evidence of whether the agent's payment was authorized), retain the evidence trail (intents, mandates, receipts -- for resolving disputes and representment), resolve liability via the mandate (within the mandate -> user authorized; outside -> the agent operator liable for the overstep; malfunction -> operator liable for the error), use representment (contesting chargebacks with the mandate evidence for legitimate payments), prevent disputes with verifiable mandates (clear authorization -- reducing and easing disputes), handle the agent-vs-human liability dimension (novel to agent payments), and manage the dispute risk -- because disputes are inevitable, and agent payments add the critical question of whether the agent was properly authorized (resolved via the mandate as verifiable evidence), making robust chargeback handling (mandate-centered, with agent-vs-human liability) essential to trustworthy agent payments.