Agentic AI Is Moving Fraud Decisions Into the Payment Itself
PYMNTS published i2c CEO Amir Wain's argument that agentic AI in payments depends on unified data, audit trails and human oversight before autonomous systems make transaction-level decisions.
Agentic AI is moving fraud management from recommendation to execution, turning what used to be a review process into something that can happen inside the transaction itself.
Amir Wain, chief executive of i2c, wrote in a PYMNTS eBook on 5 August 2026 that companies scaling these systems need guardrails in place before autonomy expands: permissioning, audit trails and human oversight.
Without that foundation, autonomy in payments is not simply an upgrade.
It becomes a source of risk.
Wain framed the shift as the third major era of AI.
Rule-based systems defined the 1980s.
Machine learning models began detecting patterns around 2010.
Agentic AI now adds perception, action and memory to the reasoning capabilities generative AI already introduced.
That matters because an agent can only act across systems it can see.
On fragmented platforms, an AI initiative quickly becomes a data-integration project.
A unified data model changes the starting point, allowing fraud systems to read signals across a customer’s full relationship and move decisions closer to the moment money moves.
Fraud management is the clearest example.
In the process Wain described, an anomaly is detected and action begins within the transaction flow.
The customer may see a quick confirmation instead of a hard decline, a dispute process or call-center review.
The point is not just automating a step, but removing it.
That change also shifts decision rights.
Judgments that once sat with analysts or service teams increasingly move to systems, while people intervene when exceptions or risk thresholds require it.
Too much oversight limits the value of autonomy.
Too little creates unnecessary risk in workflows that move money or control customer access.
Wain also reframed the build-versus-buy question.
Unless an organization operates at the scale of the largest players in its industry, building proprietary AI infrastructure from scratch is rarely the best use of capital or talent.
Companies still need to decide which capabilities truly differentiate the business and which can be handled by trusted partners.
Once autonomous systems stop advising and start acting, the test for any AI use case becomes stricter: economic benefit, structured data, real scale and auditability.
In agentic payments, those are no longer back-office checks.
They are part of the product.
