Benjamin David
Head of Intelligence, The Payments Association
As AI agents begin initiating transactions independently, new questions emerge around trust, authority and oversight.
Payment infrastructure has become faster, more connected and increasingly digital over the past decade. The industry is now confronting a new shift: agentic commerce.
While AI isn’t new to the industry, with automation and recommendation tools widespread, its role is expanding into commerce decision-making. These “agents” can compare suppliers, manage subscriptions and place orders on behalf of consumers with limited human involvement.
This shift challenges a model historically built on human initiation and oversight.
Redefining payment authority
Payment infrastructure is built around human authorisation. Transactions are initiated by individuals, with passwords, biometrics and strong customer authentication used to confirm payment intent.
With agentic commerce, this changes since consumers and businesses can delegate authority to AI systems.
That raises an important question: If authority can be delegated to AI systems, how are permissions monitored, updated or revoked? Addressing these will be important to maintaining trust and accountability.
These “agents” can compare suppliers, manage subscriptions and place orders on behalf of consumers with limited human involvement.
Existing payment infrastructures largely operate through preset rules and binary outcomes to ensure control, checks and accountability. However, the probabilistic design of agentic commerce systems fundamentally changes this model.
Entrusting payment authority to AI systems creates uncertainty about how AI-initiated transactions will be verified and supervised. Here, fraud and impersonation are concerns. What if agents are targets for payment manipulation, impersonation and financial crime?
Liability and oversight
As questions around oversight and verification magnify, liability becomes a talking point.
When a human authorises a transaction, there are generally clear lines of responsibility. If an AI system makes an incorrect or fraudulent payment, who is liable?
Concerns around oversight are also reflected in industry confidence levels. Research conducted by The Payments Association found that only 41% of UK merchants feel very confident in the current liability frameworks for AI agent transactions.1
This challenge means rethinking how responsibility is understood.
As a result, merchants, software providers, payments firms and financial institutions will need to play a role in shaping future accountability frameworks.
Ultimately, for agentic payments to scale responsibly, technological innovation must go hand in hand with clear accountability frameworks.
[1] The Payments Association. Agentic commerce in UK retail: An unresolved liability question. https://tinyurl.com/58byfe4w.
