Wallet and agents: which order can a merchant accept?
An agent purchase needs to stay connected to an identifiable order. Define what to check before accepting, repeating or escalating a transaction.
Leggi in portogheseBy Pentesty
The documented flow and the business question
Official ChatGPT Wallet documentation describes customer purchase review and approval, payment with a network token and processing through an existing checkout. The business remains responsible for the order and fraud decisions. The credential for that transaction must not be saved or reused for another purchase.
The security question is the connection between approved intent and the operation the store accepts. This is a proposed assessment, with no allegation of a product vulnerability or claim that tests have already been performed. Each business should define expected behavior using its checkout, agreements and authorized scope, keeping assumptions visible to everyone participating in acceptance.
Exercise mismatches with controlled data
In an authorized test environment, choose a valid purchase and a variation that store policy should reject. One hypothetical example changes quantity after the point the team has designated as confirmation. Agree on the expected result beforehand: the operation should stop or follow the applicable confirmation procedure, with enough observations to explain how the resulting decision was reached.
Include a repetition of the same request and a resumption after a simulated interruption. Ask the team to document how it distinguishes a repeated attempt from a legitimate new order. This article assumes no particular platform feature that solves that distinction. The exercise should assess behavior actually implemented by the merchant and participating providers, within the approved test scope.
Preserve the risk decision
Define who monitors declines and who may review an exception. Each escalation should connect the test order, the reason reported by the system and the authorized human decision. A change intended to enable the flow needs an owner, expiry and review criterion. The team should demonstrate which controls were assessed following that change and which questions remain unresolved.
Also exercise support handover: give an order identifier to someone who did not participate in the test and ask them to locate the purchase status. Record whether they can identify the next owner without seeing payment data beyond what is necessary. This evaluates work across teams, including situations where a technical result does not yet allow the order to be concluded.
Accept coverage that can be repeated
The closing record should list executed scenarios, relevant versions and configurations, and observed results. Assign exceptions to owners and dates. Distinguish what was checked in testing from what still needs an approved operational validation. A successful sample result supports only the scope represented by that sample, with remaining questions carried into the handover rather than hidden by a completion label.
Repeat relevant scenarios when checkout, the risk provider or order-confirmation rules change. Preserve earlier results as history and explain why another assessment is needed. The goal is continuity of authorization and support evidence. The proposal does not replace contractual decisions or promise to prevent every form of fraud, and acceptance should keep those practical limits visible.
Frequently asked questions
Does this analysis demonstrate a Wallet vulnerability?
No. It proposes authorized scenarios for assessing the business’s own checkout and processes. Conclusions depend on observations made during the approved assessment.
How should a mismatch be selected for testing?
Choose an alteration with the business owner that store policy should reject or reconfirm. Define the expected decision and the observations needed to verify it beforehand.
Who should participate in acceptance?
Include application, payments, risk and support owners according to scope. Agree on the evidence each receives and who decides how outstanding findings will be handled.
Can the exercise be repeated after a change?
Preserve scenarios, test data and acceptance criteria. Review which scenarios represent the altered flow before performing the next authorized assessment.
Sources and references
Original Pentesty analysis, published November 13, 2026. The proposed scenarios are editorial guidance, with scope and validation to be defined for each environment.