.png)
The Agentic Payments Split: Australia's Banks Want Rules. Card Networks Don't.
Executive summary
Australia is not debating whether AI agents can buy products. That capability is already entering live payment trials.
The question now being written into policy is more consequential: who pays when an agent makes the wrong purchase,uses the wrong payment rail or acts outside the customer’s intent?
The Reserve Bank of Australia’s 2026 Review of Payments System Regulation has exposed a clear divide. Commonwealth Bank,Westpac and Macquarie want the RBA to engage early with agentic payments because liability,market concentration,routing and platform gatekeeping risks are still unresolved. Visa,Mastercard,Apple and American Express argue that formal regulation would be premature while standards,commercial models and liability frameworks are still developing.
This is not primarily a technology decision. It is a merchant economics decision.
The rules will determine who absorbs fraud,disputes,refunds,mis-executed instructions and rising acceptance costs. Merchants are the party most directly exposed to those outcomes,yet their voice is less prominent than those of banks,networks and technology platforms.
The RBA released its Issues Paper on 25 June 2026 and published non-confidential submissions in August. It plans to publish regulatory priorities by the end of 2026,consult on those priorities in 2027 and announce conclusions in 2028. The liability architecture for agentic commerce is therefore being shaped now.
What do the RBA submissions actually reveal?
The RBA’s Issues Paper specifically asks whether AI agents create competition,efficiency or financial safety issues in e-commerce. It highlights three merchant-facing concerns: agents may restrict payment choice,add new fees and create disputes over consumer intent and liability.
The submissions sharpen that question.
CBA’s submission says agentic commerce resembles mobile wallets before their rapid adoption. Its position is that early engagement could prevent market failure later. CBA identifies risks including:
New platform gatekeepers controlling ranking,defaults and access.
Additional merchant fees,including a reported 4 per cent fee attached to ChatGPT’s Instant Checkout before the service was paused.
Agents routing transactions through higher-cost or less-safe rails.
New scam and dispute vectors caused by gaps between customer intent and an agent’s literal interpretation.
Insufficient transaction data for issuers and acquirers,including agent identity,intent mandates and purchase details.
Structural advantages for large merchants and marketplaces that can reduce token costs through scale.
CBA’s two interim “no regrets” requests are straightforward: greater fee transparency from wallet providers and early engagement between the RBA,schemes and AI platforms on agentic commerce expectations.
By contrast,Visa’s submission argues for an outcomes-focused approach and cautions against prescriptive regulation. Visa points to its Trusted Agent Protocol,Visa Intelligent Commerce and the Agentic Ready Program,which launched in Australia in April 2026 with nine financial institutions.
Visa’s argument is that merchant adoption will be governed by commercial logic: agentic commerce will scale only when incremental revenue exceeds the incremental cost of acceptance.
Both positions contain a practical truth. The networks are building private trust infrastructure quickly. Banks are asking who carries the downside when that infrastructure fails.
Who is liable when an agent mis-executes an instruction?
A customer might instruct an agent to buy a flight under A$500,from a specified airline and with checked baggage included. The agent purchases a non-refundable A$900 fare from another carrier.
Was the transaction authorised?
The answer may depend on whether the agent misunderstood the instruction,whether the mandate was ambiguous,whether the merchant’s product data was incomplete or whether the payment provider failed to transmit the relevant intent signals.
Traditional card rules were designed around a human cardholder,a merchant and a payment network. Agentic commerce introduces a software participant that can discover,select and execute. Liability cannot be allocated reliably if the payment system records only a token and an approval response.
This is the core of agentic commerce liability. A merchant needs to know whether it is responsible for the outcome simply because it fulfilled the order,even when the agent exceeded its authority.
Can issuers and acquirers see the agent’s identity and intent?
CBA says issuers,schemes and acquirers often cannot see that an agent initiated a transaction. That creates a major problem for both fraud prevention and dispute resolution.
An authenticated agent is not automatically an authorised transaction. A payment foundation model for agentic commerce needs to represent at least four relationships:
The identity of the agent.
The operator or platform controlling the agent.
The customer or business that delegated authority.
The precise limits of the mandate.
Those limits should include value,category,merchant,delivery conditions,expiry and revocation status.
Without this data,an issuer may treat legitimate automation as suspicious bot traffic. A merchant may be unable to prove that the purchase matched the customer’s instructions. A dispute team may have no reliable record of what the customer intended.
Who controls routing and payment defaults?
The RBA is already concerned about least-cost routing and merchant choice in online and mobile debit payments. Agentic commerce could intensify that issue.
An agent may select a payment method because of rewards,commercial incentives,technical integrations or platform defaults. That choice may not reflect the merchant’s preferred routing strategy or lowest cost of acceptance.
The RBA notes that agentic payments may initially be limited to international card networks. It also asks whether agents could undermine competition by routing transactions through particular networks or payment methods.
That matters more as Australia approaches 1 October 2026,when the card surcharge ban and the reduction in the interchange cap from 0.80 per cent to 0.30 per cent take effect. Merchant scrutiny of acceptance costs is already high.
For merchants,least-cost routing cannot mean simply accepting the route selected by an AI platform. It should include visibility of the route,clear commercial terms and the ability to measure cost,authorisation performance,fraud outcomes and payment success rate.
Who carries fraud,scam and dispute costs?
Visa points to strong fraud outcomes. It says overall card fraud in Australia fell by about 28 per cent between 2020 and 2025,while domestic and cross-border card-not-present fraud fell by about 40 per cent. Visa Advanced Authorisation helped Australian financial institutions prevent A$722 million in fraud in the 12 months to June 2025,and tokenisation reduced fraud rates by 58 per cent across Asia Pacific compared with non-tokenised transactions.
Those controls are valuable. They do not resolve agentic payment liability.
The new risk is not only stolen credentials. It is a legitimate customer,legitimate agent and legitimate payment being used to purchase the wrong thing.
The gap between consumer interest and merchant trust makes the problem visible.
Visa reports that 44 per cent of Australian consumers would be more likely to shop with a retailer offering an AI agent.
PYMNTS Intelligence reports that 93 per cent of merchants believe the AI provider should bear the financial loss for incorrect purchases.
In the United States,agentic commerce remains below 1 per cent of e-commerce despite a reported 4,700 per cent year-on-year increase in AI-generated retail traffic. Only 23 per cent of US consumers trust generative AI to handle payment transactions.
The message is clear: traffic is arriving faster than trust.
Why should Australia look at Europe,the US,the UK and Singapore?
Europe has chosen a regulatory-first path. Agent payment rails are live across more than 30 banks under the PSD3/PSR architecture. Payment institutions are moving from Know Your Customer towards Know Your Agent. Santander,Mastercard,ING and Worldline have executed live end-to-end payments,and an ING/Worldline/Visa transaction in Germany used Visa Payment Passkeys in July 2026.
The United States has taken a different path. Agent platforms are expanding quickly,but merchant adoption remains limited because Regulation E liability is unresolved. The AI AGENT Act focuses on fiduciary duties rather than loss allocation for agent misexecution,while the Consumer Bankers Association has called for private network rules.
The UK has already moved through the Mills Review and its AI Adoption Plan,which recommend trust frameworks,Know Your Agent protocols and legal and liability foundations. Singapore’s MAS and IMDA are also developing governance frameworks and safeguards for AI agents in finance.
Australia now sits in the decision window between those models.
What should merchants do before the RBA concludes in 2028?
Merchants should not wait for final regulation. They should use the 2026–27 window to establish their commercial position and create evidence about the risks they are willing to accept.
A 10-point merchant readiness and liability checklist
Capture intent and mandate expiry. Store the customer’s instruction,scope,approval timestamp and expiry or revocation status.
Set spend and category limits. Define maximum value,approved categories,merchant restrictions,currency and delivery conditions.
Verify agent identity. Require an auditable agent identity,operator identity and trust status before processing.
Define who carries the loss. Contractually identify responsibility for mis-execution,unauthorised purchases,fraud and agent compromise.
Document dispute and refund paths. Confirm chargeback rights,refund timelines,evidence standards and the party responsible for customer remediation.
Set routing expectations. Require visibility of the selected rail and assess least-cost routing against cost,security,approval performance and resilience.
Demand data access. Ensure the acquirer and merchant can receive agent identity,intent,mandate and transaction-level data.
Protect reconciliation. Separate agent-originated transactions in settlement,reconciliation,accounting and reporting without creating a disconnected ledger.
Prepare customer support reversals. Create a fast process for cancellations,order corrections,refunds and agent permission revocation.
Run a bounded pilot. Start with low-value categories,explicit mandates,short expiry windows and a limited customer cohort. Measure disputes,refunds,fraud,acceptance cost and payment success rate.
Quantum Payments’ unified commerce platform can help merchants connect payment orchestration,omnichannel acceptance,accounting and business intelligence rather than treating agentic payments as a separate checkout experiment. Our payment orchestration resources provide further context on routing and multi-provider control.
This is also the operational next step after our analysis of why the payment stack is ready for AI agents while merchants are not. Merchants should also review the emerging Know Your Agent trust layer and the relationship between fraud controls,approval economics and payment success rates.
What should happen next?
The RBA should treat agentic payments as a priority for early engagement,without prescribing a single technology.
The immediate objective should be a common baseline for:
Agent identity and operator traceability.
Intent and mandate evidence.
Minimum data passed to issuers,acquirers and merchants.
Dispute,refund and liability allocation.
Fee and routing transparency.
Consumer and merchant controls.
Interoperability across competing networks.
Australia does not need to choose between innovation and safety. It needs to ensure that innovation does not embed a liability model in which merchants pay for decisions they cannot see,control or challenge.
The regulatory fork is open now. By 2028,the commercial terms may already be difficult to unwind.
Frequently asked questions
What are agentic payments?
Agentic payments are transactions initiated or completed by an AI agent on behalf of a consumer or business. The agent may search,compare,select and purchase products or services within delegated rules.
Why is agentic payment liability unresolved?
Existing payment rules generally assume a human authorised the transaction. Agentic commerce introduces software that can interpret instructions and act autonomously,creating uncertainty when the outcome differs from the customer’s intent.
Do Visa and Mastercard oppose regulation?
Their submissions do not reject governance. They argue that formal,prescriptive regulation is premature while technical standards,commercial practices and liability frameworks are still maturing.
Why do Australian banks want early RBA engagement?
CBA,Westpac and Macquarie are concerned that delayed action could allow platform gatekeeping,opaque fees,limited data access and unclear loss allocation to become embedded in the market.
What should merchants record for an agent transaction?
Merchants should retain the agent and operator identity,the customer mandate,spend and category limits,expiry status,purchase details,authentication signals,selected payment rail and all refund or dispute evidence.
Should merchants wait for regulation before piloting agentic payments?
No. Merchants should begin with bounded pilots,low-value transactions,explicit consent,short-lived mandates and clear contractual allocation of fraud,dispute and mis-execution losses.
Sources
SEO metadata
Meta title: Australia’s Agentic Payments Liability Split | Quantum Payments
Meta description: Australia’s banks want early rules for agentic payments liability,while card networks argue regulation is premature. Learn what merchants should do now.
Suggested slug:australia-agentic-payments-liability-split
Primary keywords: agentic payments,AI payments,agentic commerce liability
Secondary keywords: AI payment fraud liability,payments foundation model,payment success rate
Daily handover to Sonny
Publishing details
Blog title: The Agentic Payments Split: Australia's Banks Want Rules. Card Networks Don't.
Blog URL:https://www.quantumpayments.io/post/australia-agentic-payments-liability-split
Wix slug:australia-agentic-payments-liability-split
Publishing instruction: Schedule in Wix for Tuesday 22 September 2026 at 7:30am AEST. Confirm the post is scheduled at least one hour before it goes live.
Primary angle: Australia’s 2026–27 regulatory window will decide who carries the financial loss when an AI agent mis-executes a purchase. Banks want early governance;card networks argue regulation is premature. The merchant economics decision is being made now.
Visual assets:
Relevant tags: Agentic Payments,AI Payments,Agentic Commerce,Payment Regulation,Payment Orchestration,Merchant Payments,AI Fraud,Payment Success Rate,Fintech,Unified Commerce
Recommended organisation tags: Reserve Bank of Australia,Commonwealth Bank of Australia,Westpac,Macquarie Bank,Visa,Mastercard,Apple,American Express,Quantum Payments
Hashtags: #AgenticPayments #AIPayments #AgenticCommerce #PaymentRegulation #MerchantPayments #PaymentOrchestration #AIFraud #PaymentSuccessRate #Fintech #UnifiedCommerce
LinkedIn post 1, 8:08am AEST
Angle note: Data-led and strategic. Focus on the regulatory split and the gap between consumer interest,merchant trust and actual adoption.
Exact copy:
Australia’s agentic payments rules are being written now.
The split is already clear:
CBA,Westpac and Macquarie want the RBA to act early on liability,market concentration and platform gatekeeping.
Visa,Mastercard,Apple and American Express argue formal regulation is premature.
Europe already has agent payment rails live across more than 30 banks.
In the US,agentic commerce remains below 1% of e-commerce despite a reported 4,700% year-on-year increase in AI-generated retail traffic.
The trust gap is even sharper.
44% of Australian consumers say they would be more likely to shop with a retailer offering an AI agent.
93% of merchants say the AI provider should bear the financial loss for incorrect purchases.
That is the unresolved issue:who pays when the agent gets the instruction wrong?
The answer will affect chargebacks,refunds,fraud costs,merchant acceptance fees,routing and payment success rate.
The RBA plans to publish priorities by the end of 2026,consult in 2027 and conclude the Review in 2028.
Merchants should not wait until the final rules arrive. The liability architecture is being shaped now.
Read the full analysis: https://www.quantumpayments.io/post/australia-agentic-payments-liability-split
First comment:
The merchant question is simple:can your payment stack prove what the customer authorised,which agent acted and who carries the loss when the outcome is wrong?
Relevant tags: Reserve Bank of Australia,Commonwealth Bank of Australia,Westpac,Macquarie Bank,Visa,Mastercard,Apple,American Express,Quantum Payments
LinkedIn post 2, 3:23pm AEST
Angle note: Operator-focused and materially different from the morning post. Give merchants a practical checklist they can run this week.
Exact copy:
Before piloting agentic payments,run this 10-point merchant check:
Capture the customer’s instruction and mandate expiry.
Set spend,currency,category and merchant limits.
Verify the agent and its operator.
Contractually define who carries loss for mis-execution,fraud and compromise.
Confirm dispute,chargeback and refund paths.
Make the selected payment rail and routing logic visible.
Require agent identity,intent and purchase data in the transaction record.
Reconcile agent-originated payments through the same financial control environment.
Prepare fast reversals for incorrect orders,refunds and revoked permissions.
Start with a bounded pilot using low values,short mandates and a limited customer cohort.
Do not measure the pilot only by conversion.
Track:
Payment success rate
Fraud and scam losses
Dispute and refund volumes
Cost of acceptance
Routing outcomes
Customer support reversals
Time to resolve an incorrect purchase
The key operational question is not whether an AI agent can complete checkout.
It is whether your business can observe,challenge and reverse what happens next.
Read the full merchant checklist: https://www.quantumpayments.io/post/australia-agentic-payments-liability-split
#AgenticPayments #MerchantPayments #PaymentOrchestration #AIFraud #PaymentSuccessRate #UnifiedCommerce
First comment:
If your acquirer cannot provide agent identity,intent and mandate data,add that requirement to your next provider review before approving a live pilot.
Relevant tags: Quantum Payments,retail operators,e-commerce leaders,payments teams,acquirers,merchant finance leaders