How should identity, fraud and payments teams share ownership of agentic commerce risk?

How ownership should be divided across identity, fraud, and payments

Agentic commerce works best when the three teams own different layers of the same control plane. Identity owns delegated authority, fraud owns behavioural and transaction-risk signals, and payments owns authorisation evidence, dispute readiness, and settlement outcomes. The operating model should be explicit enough that one team can answer “who allowed this action,” another can answer “why was it suspicious,” and another can answer “what evidence supports the payment decision.”

The practical test is whether each team can see the same transaction with its own lens without redefining the others’ job. Identity should not be reduced to account management, fraud should not be treated as a late payment review, and payments should not be asked to infer intent from authentication data alone. Joint ownership is about shared accountability, not shared ambiguity.

Where the handoffs need to be designed, not improvised

Delegated shopping creates a path from principal to agent to merchant, and the handoff points matter more than any single control. Identity teams should define the allowed actor, scope, duration, and revocation path for the delegation. Fraud teams should receive enough context to distinguish legitimate agentic behavior from automation abuse, while payments teams should preserve the artefacts that explain authorisation and any later dispute.

That design becomes especially important when the same agent can browse, select, and purchase across sessions or devices. The Agentic Commerce Identity Guide is useful here because it frames shopping mandates, tokenised credentials, and verifiable intent as part of the identity problem, not an afterthought. In parallel, the AI Agent Authorisation Guide helps teams separate standing permission from per-action approval so the fraud and payments layers are not forced to compensate for overbroad access.

A further practical boundary is observability. The AI Agent Observability, Audit and Incident Response Guide supports the operational view that attribution, audit trails, and revocation signals must be available when a purchase later looks abnormal. Without that evidence, fraud and payments teams end up reconstructing intent from incomplete logs.

What good shared ownership looks like in practice

Good joint ownership starts with one policy for the transaction and three team-specific decision rights. Identity decides whether the agent may act at all, fraud decides whether the behaviour fits the expected risk pattern, and payments decides whether the transaction can be authorised, reversed, or defended. The point is not to duplicate approvals, but to avoid gaps where each team assumes another team already handled the risk.

This is easier to sustain when the organisation treats the agent as a governed actor rather than a generic automation tool. Agentic AI Identity Guide is relevant because it connects delegation, lifecycle, and ownership, which are exactly the dimensions that determine whether a commerce action is attributable after the fact. For teams that need a broader control model, Zero Trust for AI Agents reinforces the discipline of verifying the principal, the request, and the action rather than trusting the session by default.

It also helps to align the process with payment proof, not just fraud score. Fraud teams often see signals that are useful for risk ranking, but payments teams need evidence that survives disputes and chargebacks. That means capture of mandate, consent, device or session linkage where appropriate, and the reason a purchase was allowed even when it was automated.

Practitioner Guidance

What to prioritise: Define one owner for delegation policy, one for abuse detection, and one for payment evidence retention. If no team can state what it would need to prove after a disputed agentic purchase, the governance model is still too vague.

Decision rule: If an action can spend money, change fulfilment, or create later dispute exposure, require a written handoff between identity, fraud, and payments that states which team can block, which can flag, and which can reverse.

What good looks like: The teams use the same transaction record but make different decisions from it, with clear escalation when the agent acted within scope but the pattern still looks suspicious, or when the pattern is acceptable but the proof is too weak.

Common mistake: Treating fraud review as a substitute for authorisation, or treating authorisation as proof that the transaction was safe. In agentic commerce, those are related but not interchangeable judgments.

Practitioner takeaway: The safest operating model is one where delegated buying is governed as a shared business process, but each team owns a distinct decision that can be audited independently.

Related resources from NHI Mgmt Group

• Why do agentic commerce flows change identity risk for merchants and IAM teams?
• How do identity and cloud teams share responsibility for agentic AI risk?
• Why do fraud teams and identity teams need shared ownership of cash-out risk?
• Why do online gaming environments create so much fraud risk for identity and payments teams?