Does agentic commerce shift fraud liability onto merchants?

Sam Boboev, writing on Medium for Fintech Wrap Up, argues that agentic commerce introduces a "fifth participant" — the AI agent — that breaks the traditional four-party card payment model and pushes risk downstream onto merchants. He writes that "when an AI agent buys the wrong product or acts without explicit consent, the merchant is the one left holding the bill," and warns that "traditional fraud detection is dead in this environment." Because risk models that rely on "monitoring typing cadence or mouse movements" have nothing to measure once a script or server-side agent is placing the order, Boboev says the disappearance of these behavioral signals will produce a spike in "friendly fraud" and "hallucination disputes" — his term for cases where an agent misreads user intent and orders the wrong item. He states that across "every protocol currently being proposed, from OpenAI to Google, the merchant remains the MoR [Merchant of Record]," so merchants absorb the cost of autonomous execution without the fraud-detection protections they historically relied on.

The mechanism he gives is a "shifting trust perimeter": payments no longer originate at a merchant's own storefront, where the merchant controls the environment and can instrument it, but inside "external LLMs, personal assistants, and procurement bots" the merchant did not build and cannot inspect. This decentralization removes device fingerprinting and behavioral biometrics as trust signals, since there is, in his words, "no typing cadence to measure when a script is making the call" and "no mouse movement when a server-side agent is interacting with an API." Liability, he argues, still attaches at the point of sale to the Merchant of Record regardless of which intermediary protocol routed the transaction.

This complements rather than overlaps with Does naming a payments brand increase trust in AI transactions?, which measures the consumer-side trust gap in letting an agent pay; Boboev's claim is the matching merchant-side exposure once that payment happens. It also sits against How can AI agents prove who authorized their actions? and the AP2 "mandates" described elsewhere in this same source: those are attempts at exactly the audit trail Boboev says merchants need, but his argument is that even a well-documented authorization record does not relieve the merchant of Merchant-of-Record liability for a disputed purchase. It also qualifies Can a protocol split agent access from agent incentives? — permissioning who can access what does not, on Boboev's account, resolve who pays when an agent gets the order wrong.

The excerpt is a single fintech commentator's analysis, not a study; it cites no fraud-loss data, chargeback-rate figures, or named merchant cases to quantify how large this liability shift actually is, and it does not address whether emerging protocols like AP2's signed mandates might eventually be recognized by card networks as evidence sufficient to shift liability back toward the agent's operator. If his reading of "every protocol" holds, merchants adopting agentic-checkout standards should expect to negotiate liability allocation explicitly rather than assume new protocols solve it by default.