The Payments Industry Inflection Point: Part 1 Agentic Payments, Trust & The Path Forward
Two forces are reshaping the payments industry simultaneously, and neither is moving slowly.
The first is agentic AI: systems that don't just assist humans but act on their behalf, initiating transactions, negotiating terms and executing payments autonomously. The second is the digitization of assets, including stablecoins and tokenized instruments, which are restructuring how value moves across borders and systems.
Together, these forces are creating what our financial services experts describe as a genuine inflection point, one that demands more than operational adjustment. It demands a rethinking of trust, identity, liability and the very architecture of financial infrastructure.
To unpack what this means in practice, we spoke with two of EPAM's leading voices in payments: Chinmay Jain, VP Client Partner, Financial Services and Michael Nelhams, Head of Open Banking and Payments.
On Why Today's Disruption Is Different
Q: The payments industry has always been in motion, but the pace of change feels different right now. What's driving this moment, and why does it feel more significant than previous waves of disruption?
Chinmay Jain: We're at an inflection point in the history of payments. I'd call it a multi-vector convergence taking place, with two very important forces impacting payments simultaneously: agentic commerce and the digitization of assets and stablecoins. We're also seeing this reflected in research. Data from EPAM's 2026 Consumer Banking Report, which surveyed 14,000 banking customers across 11 countries, shows that more than 50% of consumers are now comfortable with AI in financial services. Among Millennials and Gen Z, that figure rises to more than 60%.
What's happening is a migration away from using AI in back-office fraud decisioning systems toward using AI for payment initiation, with entire transactions being orchestrated by AI. That shift, coupled with the movement around digital assets and stablecoins, is creating a genuine inflection point that could upend the architecture of infrastructure for payment companies.
Michael Nelhams: Over the previous decade, the industry has been mainly focused on transforming the speed of national payments, implementing infrastructure and processes in support of instant payments, whilst also opening up the once secretive bank held data to meet open banking requirements. This has been broadly implemented at national levels.
We are now seeing a new wave of thinking focused on the antiquated cross border infrastructure, moving past DLT experimentation, with the parallel rise of AI which is shaking up the global technology industry.
DLT solutions incorporating asset tokenization are proving capable of enabling fully auditable, cross-border asset exchanges backed by payments products. For banks and asset management firms, these solutions can reduce processing costs while improving transaction traceability — addressing the opacity that still exists within correspondent banking networks. In parallel, AI brings both new intelligence opportunities to accelerate decision-making and improve operational efficiency, while reducing reliance on complex rule sets and enabling faster technology change. The whole industry is frantically jumping in this bandwagon to gain a competitive advantage.
On Trust & Authentication Challenges Surrounding Agentic Payments
Q: When an AI agent can browse, compare, negotiate and complete a purchase on a consumer's behalf, the entire payments infrastructure has to adapt. What are the most pressing challenges that need to be solved before agentic payments can truly scale?
Chinmay Jain: The biggest challenge is that current payment infrastructure was built for a human in the loop. Identity verification relies on SMS one-time passwords, biometrics and similar mechanisms. Agentic commerce removes the human from the loop entirely. We're moving from a "know your customer" model to a "know your agent" model, and the questions become fundamental: How do you authenticate something that isn't human? How do you establish an agent's identity? How do you verify intent and confirm that authority has been properly delegated? The entire chain has to be rebuilt around trust and identity at every step.
Michael Nelhams: The rapid development of agentic commerce is creating complexity across the ecosystem as competing solutions, many unproven, are being experimented with and implemented into production. We're seeing three distinct layers of agentic solutions emerging, who will fight it out over the coming years. The first consists of cross-industry AI providers, such as Anthropic, Google and OpenAI, offering general-purpose solutions ideal for aggregators. The second includes specialist commerce agents built for specific industries with more precise tuning to consumer intent. The third is made up of proprietary agents developed by individual enterprises using merchant-specific data and ecosystems. I think corporates who jump on the general solutions are by default having to open up their proprietary data to the big tech companies.
All three groups of players are building towards the same general goal, however, on different protocols, which means interoperability is limited right now. Trust also varies significantly by transaction type. For high-value, one-time transactions like mortgages or property purchases, consumers are unlikely to delegate to an agent in the near term. For lower-value, repeating transactions like subscription management or utility switching, trust comes more quickly. Adoption will build from low-value eCommerce upward.
On Security, Fraud & Authentication
Q: What does all of this mean for fraud and security?
Chinmay Jain: The authentication gap is one of the most serious near-term risks. Existing fraud detection systems were built around human behavioral patterns. Agents behave differently, and fraudulent agents can be difficult to distinguish from legitimate ones. This requires AI-based fraud decisioning to be deeply embedded, not bolted on, with continuous behavioral monitoring throughout the lifecycle of a transaction.
Michael Nelhams: The fraud challenge is real and underappreciated. Every week we hear about agents that have “gone rogue,” hacking businesses when the AI firms “lost control” of the agent. If this is the AI firms losing control, then bad actors will certainly look to enhance these behaviors. I think edge cases that traditionally may not have been tested, could be easily be exploited by AI agents, putting a much higher focus and by implication cost on cyber security.
Without robust mechanisms to distinguish authorized agent activity from unauthorized or fraudulent activity, losses will grow as agentic volume grows. Insurers of cyberattack losses will need to rethink their entire model and fast.
One emerging approach is the "know your agent" framework, analogous to know your customer, which establishes verifiable identity and authorization scope for agents. This requires cryptographic identifiers linked back to verified humans or legal entities, dynamic trust scores based on behavioral data and immutable audit trails. None of this infrastructure exists at scale today, which is why fraud remains one of the structural barriers to broader agentic commerce adoption. The equivalent agreement to implement Open Banking APIs with a secure trust framework to enable the sharing of data took many years and is still being tweaked.
On Agentic Payments Responsibility
Q: When an AI agent executes a transaction that goes wrong, whether that's a fraud claim, a chargeback or a misinterpretation of consumer intent, who is responsible? How does the current regulatory framework hold up?
Chinmay Jain: In agentic commerce, multiple agents could participate in a single payment lifecycle and the decisions they make are probabilistic. If there's a fraudulent transaction, how do you determine if it's the consumer's responsibility, the agent provider's, the AI platform's, the merchant's or the issuer's? It's comparable to the autonomous vehicle question: If a self-driving car causes an accident, who is liable? The answer isn't clear. For agentic payments to scale, this has to be resolved. One direction worth exploring is to build dispute resolution frameworks the way we already do for human-initiated payments, paired with insurance mechanisms made specifically for AI-initiated transactions.
Michael Nelhams: In the U.K., banks are currently liable for most fraudulent or lost payments under a range of regulations, including the Payment Services Regulations, the PSR's mandatory APP fraud reimbursement rules and Financial Ombudsman Service oversight. This legislation wasn't written with agentic payments in mind. In Europe, PSD3 includes some liability provisions, but agentic payments aren't explicitly covered there either.
The EU AI Act and the proposed Financial Data Access Regulation both touch on areas where agents will operate, but applying them to agentic payments requires interpretation that hasn't happened yet. Following the logic of autonomous vehicle law, liability may ultimately fall to whoever owns the agent. Governments and regulators tend to move slowly on these questions, and banks tend to end up holding responsibility. The next 12 to 24 months will likely bring clearer legislative direction, but for now it's unresolved. And that resolution is a precondition for real scale.
What Comes Next
This conversation is the first in a two-part series. In Part Two, Chinmay and Michael turn to the second major force reshaping payments: stablecoins and digital assets. They examine how programmable money changes settlement, what core system modernization actually requires and where financial institutions should be placing their strategic bets right now.
