Brief IA
›

Visa, BILL, and Billtrust Test AI Agent Payments

AI Tools·Tom Levy·

Visa, BILL, and Billtrust Test AI Agent Payments

Visa, BILL, and Billtrust Test AI Agent Payments
Key Takeaways
1Visa, BILL, and Billtrust launched an AI agent payment pilot in October.
2The agents settle already approved invoices without choosing which ones to pay.
3A payment trail does not necessarily identify the person who authorized the agent.
Why it matters : the pilot entrusts agents with executing payments for businesses, but not the decision to select invoices.

Visa, BILL, and Billtrust launched a pilot program in October allowing authenticated AI agents to settle invoices approved by buyers. These agents can act on behalf of participating companies without selecting the invoices to be settled themselves. The authorization of a payment remains distinct from the mere technical ability to execute it.

The Payment Trail Does Not Prove Authorization

Recording a transaction performed by an agent does not necessarily demonstrate that it was authorized. The payment trail establishes its execution without necessarily identifying the person who authorized the agent. Proof of authorization must clarify the basis on which the transaction was permitted.

To establish this chain, the elements to retain include the identity of the agent, the accesses actually used, the transaction in question, the applicable rule, and the validations obtained across the systems traversed. This must allow for the identification of the person who entrusted the task. For IT departments, this work must be conducted with financial, legal, and security services before allowing agents to carry out transactions.

Contracts Can Limit Supplier Liability

In the tested setup, payments circulate between the systems used on the buyer's side and the supplier's side. This case marks the beginning of delegating commercial transactions to agents: until now, employees had to navigate applications, validate requests, and then proceed with payments.

For Veronica Glick, a partner in cybersecurity and data protection at Mayer Brown, technical protections do not exhaust the topic. She recommends that IT leaders and legal professionals clarify how responsibilities are distributed between the client company and its AI providers if an agent exceeds their mandate. Contracts can delineate permitted actions, the systems involved, and cases requiring human validation, to clarify both the authorizations granted by the client and the limits of the supplier's liability.

La lettre du soirGratuit · 19 h

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Managing an error is also part of the provisions to be established before any incident. This involves designating those responsible for investigations, cancellations, and reporting the issue to the relevant parties.

Technical Access Does Not Equate to Mandate

Vibhuti Sinha, product director at Saviynt, a specialist in identity security, distinguishes three elements: the identity specific to the agent, the applications accessible to them, and the actions they are authorized to perform. He cites a hypothetical situation where technical access would allow an agent to validate an invoice reaching $2 million. According to him, the legitimacy of this validation depends on the delegator, the action contemplated, and the business context at that moment. The transition between multiple applications complicates this distinction, as they do not all assign the same meaning to access.

Defining the mandate thus requires identifying the person or business function that grants it, and then specifying the permitted operations, the systems involved, the recipients, and the financial limits. Thresholds such as exceptions must determine when a person reviews the action before its execution. Veronica Glick recommends proportioning these validations to the consequences of the act and the possibility of reversing it.

Brief IA — L'actualité IA en français

L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.