Payments Agent Governance Compliance | Moring
Compliance for payments agent governance means mapping each agent and payment workflow to the regimes that govern it: FFIEC examination expectations in the United States, DORA for EU operational resilience, the EU AI Act for systems used in Europe, and model risk management guidance including SR 26-2. The compliance posture rests on evidence: inventories, scoped permissions, approval records, and reconstructable audit trails.
Short answer
Compliance for payments agent governance means mapping each agent and payment workflow to the regimes that govern it: FFIEC examination expectations in the United States, DORA for EU operational resilience, the EU AI Act for systems used in Europe, and model risk management guidance including SR 26-2. The compliance posture rests on evidence: inventories, scoped permissions, approval records, and reconstructable audit trails.
How it works
Compliance for agentic payments starts with an honest inventory. Supervisors consistently find that institutions use more AI than any single team can see, because agents arrive embedded in vendor platforms, internal tools, and third-party services. Examination preparation research notes that many institutions use AI through capabilities embedded within existing systems and third-party platforms, often with little visible change to the end user. A payments organization that cannot enumerate its agents, their workflows, and their permissions cannot demonstrate compliance with any regime, because every framework begins by asking what systems exist and who owns them.
In the United States, the applicable structure is examination expectation rather than agent-specific rule. The Federal Financial Institutions Examination Council has not issued a dedicated AI examination booklet, but its existing IT handbook expectations, including governance, audit, and operations booklets, apply to agent-driven changes in payment systems, and examination preparation guidance recommends applying those existing expectations to emerging AI risks rather than assuming agents fall outside review. For model risk, the revised guidance issued in April 2026 as SR 26-2, alongside a companion supervisory bulletin, updated expectations for model risk management at financial institutions, and payments organizations should map their agent-driven scoring and decisioning to that guidance where it applies.
For EU operations, two regimes dominate. The Digital Operational Resilience Act, Regulation (EU) 2022/2554 applicable since January 2025, governs the operational resilience of ICT-supported functions at EU financial entities, and an agent that changes records or moves payments is one in every practical sense. Its five obligations that reach agent deployments are ICT risk management, logging and event reconstruction, resilience testing, third-party risk, and incident reporting. Separately, the EU AI Act, Regulation (EU) 2024/1689, classifies AI systems by risk tier and phases in obligations for high-risk systems through the coming years, with obligations for general-purpose models already applying since August 2025. Payments organizations should classify their agentic systems against those tiers and document the determination.
The practical compliance architecture is the same across regimes: controls that produce evidence as a byproduct of operation. Supervisory evidence checklists developed for banking pair each governance step, including inventory, risk rating, pre-use testing, human review, ongoing monitoring, and third-party oversight, with the record that proves the step was done. Payments organizations should adopt the same pairing for agents: each workflow carries its classification decision, its permission scope, its approval thresholds, its test results, and its monitoring plan, each stored as an artifact rather than asserted in policy. Examiners and auditors evaluate records, not intentions.
Third-party risk deserves explicit attention because most payments agents depend on external model providers and tool vendors. DORA makes third-party risk management an explicit obligation for EU financial entities, and US examination expectations apply the same discipline to vendor-supported functions. The compliance questions are concrete: what contractual terms govern the model provider's behavior and data handling, how does the institution receive notice of model changes, what exit provisions exist, and who assesses the agent platform itself as a vendor. A governance program that treats its own tooling as exempt from vendor diligence fails the standard it applies elsewhere.
Compliance also has a temporal dimension that governance must absorb: obligations phase in and supervisory interpretation evolves, so the mapping between payment workflows and regimes needs a review cadence. Analysis of global agent governance notes that organizational controls rather than law will govern agents in high-stakes domains for now where rules are delayed, which places the burden of diligence on internal frameworks. Payments organizations should treat their compliance mapping as a maintained document with owners and review dates, not a one-time assessment.
The end state is a payments agent program where compliance is demonstrated from operational records: the inventory answers what exists, the classification answers what applies, the runtime controls answer what prevented harm, and the audit trails answer what happened. Organizations that reach this posture scale agentic payments automation without friction, because every new agent inherits a compliance process that produces its own evidence.
Frequently asked questions
Which regulatory regimes apply to agent governance in payments?
It depends on footprint. United States payments organizations face examination under existing FFIEC expectations, which apply to agent-driven changes even without a dedicated AI booklet, plus model risk management guidance including the revised SR 26-2 issued in April 2026. EU operations add the Digital Operational Resilience Act, applicable since January 2025, and the EU AI Act's risk-tier obligations, phasing in through the coming years. The mapping exercise, workflow by workflow, is itself the first compliance control.
How does DORA apply to payments agents specifically?
DORA never names AI; it governs ICT-supported functions at EU financial entities, and an agent that changes records or moves payments is one in every practical sense. Five obligations reach agent deployments: ICT risk management, logging and event reconstruction, resilience testing, third-party risk, and incident reporting. The translation for payments is per-agent identity attribution on every write, append-only logs that support reconstruction, a rehearsed stop drill, contractual terms with model providers, and incident evidence that assembles quickly.
What evidence do examiners expect for payments agent compliance?
Evidence pairing: each governance step accompanied by the record that proves it was performed. Practical checklists for institutions pair inventory, risk rating, pre-use testing, human review, ongoing monitoring, and third-party oversight with their corresponding artifacts. For payments agents that means classification decisions on file, permission scopes configured and logged, approval thresholds with recorded decisions, test results before deployment, and audit trails attributed to agent identities. Examiners evaluate produced records rather than policy assertions.
How should payments organizations handle vendor risk for model providers?
Extend existing third-party oversight to model and agent platform providers rather than treating them as exempt. DORA makes third-party risk an explicit obligation for EU financial entities, and US examination expectations apply vendor discipline to supported functions. The concrete items are contractual terms covering data use and model changes, incident notice commitments, audit rights, exit provisions, and a risk assessment of the governance platform itself. Payments agents amplify vendor exposure because model changes can alter transaction behavior directly.