By Nick Hossack in RBA — 08 Oct 2026

Will the RBA regulate the role of AI in making payments?

Artificial intelligence may soon do more than recommend what you buy. It could make the payment as well.

This emerging system is called agentic commerce. A consumer could tell an AI agent to find the cheapest flight to Melbourne, choose an acceptable departure time and complete the purchase — without the consumer manually visiting websites or entering payment details.

The Reserve Bank is now considering what that could mean for Australia's payments system.

Its latest payments review identifies four potential problems.

First is competition. An AI agent may choose which card network or payment provider processes a transaction. If commercial arrangements cause an agent to favour Visa, Mastercard or a particular payments provider, merchants could lose some ability to choose cheaper payment routes. That could undermine policies such as least-cost routing.

Second is cost. AI inserts another intermediary between the consumer and merchant. One submission cited reports that an AI-enabled purchasing channel had imposed an additional 4% merchant fee. The RBA has not established that such fees will become standard, but the example illustrates the risk of another payments layer adding costs.

Third is liability. Suppose you authorise an AI agent to spend $500 but it spends $800. Who bears the loss — the customer, bank, merchant, card network or AI provider? Existing chargeback arrangements were not designed for autonomous software making purchasing decisions.

Fourth is security. An AI agent could be manipulated, hacked or impersonated and then initiate unauthorised transactions.

Importantly, industry submissions generally advised the RBA not to regulate yet. Agentic commerce remains immature, evidence of actual harm is limited and premature rules could inhibit useful innovation. Stakeholders instead favoured monitoring and industry standards covering authority, transaction identification and liability.

The RBA will announce its regulatory priorities by the end of 2026.

The underlying question is new but fundamental: when software starts spending our money, who decides what it is allowed to do — and who pays when it gets the decision wrong?

BENCHMARK ANALYTICS | PAYMENTS

When AI starts making the payment

Agentic commerce could allow software to choose the product, payment provider and payment method on a customer's behalf.

1. Consumer gives AI instructions

“Buy a flight under $500 leaving after 5pm”

↓

2. AI searches and chooses

Product • merchant • price

↓

3. AI selects how to pay

Card network • wallet • payment provider

↓

4. Transaction is completed

Potentially with little or no additional human involvement

FOUR QUESTIONS RAISED WITH THE RBA

1. COMPETITION

Could an agent favour a particular card network or payment provider?

2. COST

Does AI create another intermediary charging merchants for a sale?

3. LIABILITY

Who pays if the agent exceeds the authority given by its owner?

4. SECURITY

What happens if an agent is hacked, manipulated or impersonated?

EARLY WARNING ON COSTS

Additional merchant fee cited by one stakeholder from reports of an early-2026 AI purchasing model. This is an example raised in consultation — not an RBA estimate of future AI payment costs.

REGULATE NOW?

Industry view: not yet

Stakeholders generally favoured monitoring and industry standards while agentic commerce remains at an early stage.

The policy dilemma: regulate too early and innovation may be constrained. Wait too long and international platforms may establish payment rules that become difficult to unwind.

Source: Reserve Bank of Australia, Review of Payments System Regulation, Summary of Submissions, October 2026. The RBA has not decided to regulate agentic payments and will announce its regulatory priorities by end-2026.

Never miss new Benchmark analysis