Introduction
Agentic payments are emerging as a significant issue for financial institutions as artificial intelligence moves from assisting users with financial decisions toward systems capable of initiating and managing transactions. The shift raises questions that extend beyond payment technology, including identity, consent, liability, fraud prevention and governance.
The issue is gaining attention in the UK financial sector. HM Treasury’s Financial Services AI Adoption Plan, published on 14 July 2026, identifies agentic payments as one of 10 recommendations for accelerating safe AI adoption in financial services. The plan specifically highlights uncertainty around legal and regulatory accountability, consent, liability and fraud in automated payment flows. GOV.UK
The development also illustrates a broader problem in fintech: technologies can be technically viable before the surrounding legal, commercial and regulatory infrastructure is ready for large-scale deployment. Anna Wallace, CEO of the Centre for Finance, Innovation and Technology (CFIT), argues that moving innovations from pilots into the market often requires coordination between banks, technology companies, regulators, government and standards bodies. Pasted text
What Are Agentic Payments?
Agentic payments refer to transactions in which an AI system or autonomous software agent can perform payment-related actions on behalf of a person or organisation.
Traditional digital payments generally require a customer to initiate or explicitly approve a transaction. An agentic model can introduce an additional layer in which software interprets instructions, evaluates available options and potentially executes a transaction within delegated authority.
The distinction is important. An AI assistant that recommends a product is different from an AI agent that can purchase that product and initiate payment.
That change introduces new participants and responsibilities into the payment chain. A transaction could involve a consumer, an AI agent, a merchant, a bank, a payment provider and an identity or authentication service.
HM Treasury’s 2026 AI Adoption Plan describes agentic payments as a practical near-term example of a wider class of autonomous financial systems and says existing legal and regulatory frameworks may not yet provide sufficient clarity about responsibility and consumer protection in these arrangements. GOV.UK
How Agentic Payments Could Work
The basic process involves several stages.
1. User delegation
A customer gives an AI agent authority to perform a defined task. The authority could involve parameters such as the type of purchase, spending limit or duration.
This creates an important distinction between instruction and delegated authority. A system may need to demonstrate not only that a transaction occurred, but that the agent was authorised to initiate it.
2. Agent identification
The payment ecosystem needs to establish which software agent is acting, who controls it and what authority has been granted.
This is different from conventional customer authentication because the immediate actor may be software rather than the account holder.
3. Transaction execution
The agent interacts with merchants and financial infrastructure to complete the transaction.
For this to operate across different institutions, technical standards may be required so that banks, merchants, payment providers and AI systems can recognise and authenticate one another.
4. Monitoring and dispute handling
The system must also determine what happens when a transaction is fraudulent, unauthorised, incorrectly executed or disputed.
These questions become more complicated when several autonomous systems are involved.
Why Agentic Payments Raise New Regulatory Questions
The central challenge is not simply whether AI can execute payments. It is whether the financial system can establish trust around autonomous transactions.
HM Treasury’s Financial Services AI Adoption Plan identifies three areas for an agentic-payment trust framework: legal and liability arrangements, “Know Your Agent” protocols, and authentication and governance standards. GOV.UK
Legal responsibility and liability
A conventional payment dispute generally involves identifiable parties and established contractual relationships.
With agentic payments, responsibility could become less straightforward. Questions include whether the customer, AI provider, bank, merchant or another intermediary should bear responsibility when an autonomous system makes an error.
Clear allocation of liability is therefore likely to be an important component of any scalable framework.
Know Your Agent
Financial institutions already use customer identification and authentication mechanisms. Agentic systems introduce a related but distinct question: how should an institution establish the identity and authority of an AI agent?
The 2026 AI Adoption Plan proposes “Know Your Agent” protocols as part of a potential framework for agentic payments. GOV.UK
Authentication and governance
Authentication must establish more than the identity of an account holder. It may also need to establish whether a particular software agent is authorised to perform a specific action.
Governance becomes relevant because an agent’s authority could change over time. Systems may therefore need mechanisms for granting, limiting, monitoring and revoking permissions.
Fraud and Digital Identity Are Closely Connected
Fraud is one of the principal limitations surrounding autonomous payments.
Wallace identifies deepfakes, synthetic identities and increasingly automated transactions as areas that could make financial fraud more sophisticated and scalable. She argues that trusted digital identity will become increasingly important as financial interactions become more automated. Pasted text
The risk is not limited to traditional account takeover. An autonomous system could potentially be manipulated through fraudulent instructions, compromised credentials or malicious interactions with other software.
That means fraud controls designed around human behaviour may need to operate alongside controls designed for machine-to-machine transactions.
The challenge is particularly relevant because AI adoption is already expanding across financial services. HM Treasury’s 2026 plan cites FCA and Bank of England survey findings indicating AI adoption of around 75% among surveyed financial services firms, compared with 21% across financial and real estate firms and 16% across the economy in an earlier 2025 survey. GOV.UK
From Fintech Pilots to Financial Infrastructure
Agentic payments also highlight a recurring problem in financial technology: the distance between demonstrating that something works and deploying it across a regulated market.
CFIT was created following the 2021 Kalifa Review of UK FinTech, which recommended a Centre for Finance, Innovation and Technology to strengthen coordination across the sector. HM Treasury subsequently allocated £5 million in Spending Review 2021 to establish CFIT. GOV.UK
CFIT’s approach has involved time-bound coalitions bringing together financial institutions, technology companies, government, regulators and other organisations. Its work has included Open Finance, Digital Company ID, SME access to finance, Open Property and Supported Payments. Pasted text
Wallace says one of the main barriers to fintech adoption is that the organisations required to make a system work often operate independently. Standards, governance, liability and commercial incentives can prevent technically viable ideas from progressing beyond pilots. Pasted text
This issue is particularly relevant to agentic payments because the technology crosses traditional institutional boundaries.
The Wider Implications for Banks and Fintech Companies
Banks
Banks could remain central to agentic payments because they provide regulated accounts, payment infrastructure and existing fraud and risk-management systems.
However, autonomous transactions could require banks to reconsider how they authenticate customers, monitor transactions and establish delegated authority.
Fintech companies
Fintech firms could provide the software, identity systems, payment interfaces and agent infrastructure connecting consumers with financial institutions.
Their role could therefore expand beyond traditional payment applications toward infrastructure that allows software agents to interact with regulated financial systems.
Regulators and government
Regulators face the challenge of applying existing consumer-protection, operational-resilience and financial-crime requirements to systems that may behave differently from conventional financial-service interfaces.
HM Treasury’s 2026 plan calls for continued coordination between government, regulators and industry and recommends that the upcoming modernisation of payment-services regulation address the systemic issues associated with agentic payments. GOV.UK
Risks and Limitations
The development of agentic payments remains subject to several unresolved issues.
Consent: A system needs a reliable way to establish whether a customer authorised a transaction and whether the agent remained within that authority.
Liability: Disputes may become more complex when multiple technology and financial-service providers participate in an automated transaction.
Fraud: AI can potentially increase the speed and scale of fraudulent activity as well as improve legitimate fraud detection.
Interoperability: Different banks, merchants, payment systems and AI providers may use different technical standards.
Operational resilience: Greater dependence on AI and third-party technology can introduce additional technology and concentration risks. HM Treasury’s plan specifically recommends attention to critical AI and cloud providers as AI adoption increases. GOV.UK
Regulatory coordination: Agentic commerce can cross the boundaries of financial regulation, data governance, technology regulation and consumer protection.
These limitations mean that technical capability alone is unlikely to determine the pace of adoption.
Future Outlook for Agentic Payments
The UK’s policy discussion is moving toward establishing the infrastructure required for controlled deployment rather than treating agentic payments solely as an emerging consumer feature.
HM Treasury’s Financial Services AI Adoption Plan recommends using a trust framework based on legal and liability rules, Know Your Agent protocols, and authentication and governance standards. The government has accepted the recommendations for government and said it will work with regulators and industry on next steps. GOV.UK
The broader direction is consistent with a shift in financial-technology policy from experimentation toward implementation. Wallace similarly describes CFIT’s focus as moving beyond recommendations toward developing, building and testing practical solutions that can generate evidence for wider adoption. Pasted text
The timing of that transition remains dependent on regulatory clarity, industry standards, fraud controls and the ability of different participants to establish common infrastructure.
Conclusion
Agentic payments represent a shift from digital systems that help people make financial decisions toward software that can potentially act on their instructions and transact independently within defined authority.
The technology creates opportunities for more automated commerce, but it also introduces questions around identity, consent, liability, fraud, authentication and interoperability. The UK’s 2026 Financial Services AI Adoption Plan places these issues within a broader framework for responsible AI adoption and identifies agentic payments as a specific area requiring coordinated action. GOV.UK
The development of the sector will therefore depend not only on advances in artificial intelligence, but also on whether financial institutions, technology companies, regulators and governments can establish common standards that make autonomous transactions identifiable, accountable and resilient.
