Agentic commerce is forcing marketplaces to rethink identity and risk

At a glance

What this is: This is a marketplace-focused analysis of agentic commerce, showing that AI agents change product discovery, checkout and fraud patterns by weakening human browsing assumptions.

Why it matters: It matters because marketplaces now need to govern delegated identity, seller trust and payout controls alongside customer authentication, fraud and abuse detection.

By the numbers:

• By 2030, AI agents are expected to influence up to $3 trillion in global revenue.

👉 Read Signifyd's analysis of agentic commerce for marketplaces and fraud controls

Context

Agentic commerce is the shift from human-led browsing to AI agents comparing products, narrowing choices and sometimes checking out on a person's behalf. For marketplaces, that creates an identity and governance problem as much as a fraud problem, because the platform must now interpret delegated intent, agent trust and buyer accountability in the same transaction flow.

The marketplace model adds complexity that single-brand retailers do not face. Seller-provided data, payouts, refund abuse and review manipulation all create an identity and trust boundary that agents can amplify, especially when short sessions and fewer clicks no longer map cleanly to risk. That makes this an agentic commerce and identity governance issue, not just a checkout optimisation topic.

Key questions

Q: How should marketplaces handle delegated AI agents at checkout?

A: Marketplaces should treat delegated AI checkout as a distinct trust scenario, not a normal human session with faster timing. That means validating agent authority, preserving buyer accountability and passing agent context into fraud and order systems. The goal is to decide whether the agent is authorised and whether the transaction matches expected behaviour, not simply whether a person logged in.

Q: Why do agent-led transactions break traditional fraud models?

A: Traditional fraud models rely on human browsing signals such as long sessions, broad page paths and repeated comparison behaviour. Agent-led transactions compress those signals because the agent arrives with intent already formed and often moves directly to checkout. Teams need stronger identity and relationship signals, not heavier reliance on clickstream heuristics.

Q: What do marketplaces get wrong about seller verification?

A: The common mistake is treating seller verification as a one-time onboarding step. In practice, fraud often appears later through reused bank details, repeated refunds, storefront cloning and other connected behaviour. Marketplaces need continuous seller identity monitoring that extends through payouts and post-onboarding activity.

Q: How do security teams reduce connected abuse in marketplaces?

A: They should correlate buyers, sellers, payouts and refund behaviour instead of reviewing each account in isolation. Connected abuse becomes visible when the same financial details, device patterns or transaction loops appear across multiple accounts. Relationship analysis is more effective than single-event rules when fraud is designed to look distributed.

Technical breakdown

Agent-led customer journeys and delegated intent

In agentic commerce, the agent often arrives with the purchase decision already constrained by a prompt or upstream policy, then navigates directly toward a product or checkout path. That removes many of the behavioural signals fraud models were built around, such as long browsing sessions, broad page traversal and repeated product comparison. The platform must therefore understand not only who the buyer is, but whether the agent is acting within delegated authority and whether the session context is consistent with that authority. For marketplaces, that also means distinguishing legitimate fast-path purchases from automated abuse.

Practical implication: redesign risk scoring to evaluate delegated intent and session context, not just browsing depth.

Seller verification, payout controls and connected abuse

Marketplaces sit between buyers and sellers, which means fraud can originate from either side and then spread across the platform. An abusive seller can use agent-driven buyer accounts, repeated refund requests or coordinated storefronts to mask ownership links and cash out through payout rails. This is not only an account problem, but an identity lifecycle problem across seller onboarding, ongoing verification and payout governance. KYB checks help at enrolment, but networked abuse often appears later through reused bank details, account changes and clustered transaction behaviour.

Practical implication: extend seller identity checks beyond onboarding and watch payout links for reused financial identifiers.

Catalog structure and discovery signals for AI agents

AI agents need machine-readable product data to compare and rank listings reliably. Freeform titles, inconsistent attributes and vague descriptions increase the chance that good listings disappear from discovery while low-quality or deceptive listings rise. Structured catalog fields, attribute constraints and listing validation help create a cleaner discovery layer for agents. In marketplace terms, catalog governance becomes part of identity governance because the platform is controlling what the agent is allowed to 'see' as trustworthy product data, and that affects both buyer trust and seller fairness.

Practical implication: standardise product attributes so agents evaluate comparable listings on consistent fields.

Threat narrative

Attacker objective: The attacker wants to convert delegated automation into harder-to-detect fraud, abuse payouts and manipulate marketplace trust signals at scale.

1. Entry occurs when an AI agent is given delegated shopping authority and enters the marketplace with a short, intent-driven session rather than human-style browsing.
2. Escalation happens when sellers or fraud operators exploit weak identity signals, reused payout details or automated buyer accounts to disguise connected activity.
3. Impact follows when marketplaces approve fraudulent purchases, refunds or payouts that look legitimate in isolation but are part of a coordinated abuse pattern.

NHI Mgmt Group analysis

Delegated commerce creates a new identity boundary: marketplaces are no longer deciding only whether a human buyer is trusted. They are also deciding whether an agent has permission to act, how much authority it has and whether downstream fraud controls can still interpret the transaction correctly. That shifts the governance problem from authentication alone to delegated intent, accountable identity and session-level trust. Practitioners should treat agent-originated checkout as a distinct identity event, not just another payment path.

Marketplaces need to think about seller identity as a lifecycle, not a checkpoint: the article correctly separates onboarding from ongoing monitoring, and that distinction matters. Fraud often appears after verification through reused bank accounts, repeated refunds or multiple storefronts linked to the same actor. That is a lifecycle failure, not a one-time verification failure. Practitioners should align seller governance to continuous identity assurance, not just initial KYB review.

Catalog quality is now a trust control: when agents discover products programmatically, poor metadata is not merely an operational inconvenience. It becomes a ranking, trust and fairness issue because agents depend on structured attributes to decide what is comparable and credible. In marketplace governance terms, the catalog is part of the control plane for agentic commerce. Practitioners should treat listing structure and attribute hygiene as security-adjacent controls, not just merchandising work.

Connected-risk detection will matter more than single-event fraud rules: agent-led abuse is designed to make related activity look independent. That means isolated review rules will miss the pattern even when each transaction looks plausible. The more valuable control is networked correlation across buyers, sellers, payouts and refund behaviour. Practitioners should expect the next phase of fraud governance to depend on relationship analysis rather than session heuristics alone.

Agentic commerce exposes the gap between human assumptions and machine execution: older fraud logic expects visible exploration, delayed intent and predictable checkout friction. Agents collapse those assumptions by moving quickly and directly, which makes identity and risk signals harder to interpret. This is where NHI thinking becomes useful: the system is not only handling a user, it is handling a delegated non-human actor. Practitioners should reframe the issue as machine-mediated trust rather than browser behaviour.

What this signals

Delegated commerce will push identity teams to expand their threat model beyond login events: the control question is no longer only whether the account holder authenticated, but whether an AI agent is operating within approved scope. That makes delegated authority, session context and downstream accountability central to marketplace risk governance, especially where payments and refunds are involved.

Marketplaces should expect fraud engineering to move toward relationship-based abuse rather than obvious account takeover. That means the detection problem shifts from isolated anomalies to connected patterns across storefronts, payout rails and buyer identities, which is exactly where lifecycle-aware identity controls and network analysis start to overlap.

For practitioners

• Standardise catalog attributes for agent discovery Replace freeform listing fields with structured attributes such as size, colour, material, condition, compatibility and shipping speed so agents can compare listings consistently. Apply category templates and validation rules before listings go live.
• Extend seller verification beyond onboarding Use KYB, identity verification and bank account ownership checks at onboarding, then keep monitoring for reused payout details, sudden listing changes and repeated refund requests under $10. That is where connected abuse usually becomes visible.
• Rebuild fraud models around delegated sessions Audit any rule that depends on browsing depth, page clicks or session length, then replace it with stronger signals such as device consistency, order value patterns and account relationship analysis across buyers and sellers.
• Harden checkout for authorised agents Simplify redirects, remove brittle form steps and make sure payment and order confirmation flows handle shorter, direct purchase paths without rejecting legitimate delegated transactions. Pass agent metadata downstream so fraud teams can evaluate the right context.
• Improve post-purchase communication Send explicit confirmations that include product, seller, price, shipping method and delivery window. Clear messaging reduces chargeback risk when a buyer later does not recognise an agent-placed purchase.

Key takeaways

• Agentic commerce changes the trust model for marketplaces by introducing delegated AI actors into discovery, checkout and fraud decisions.
• Seller verification, payout control and catalog governance now function as identity-adjacent controls because they shape what agents can trust and how abuse spreads.
• Teams should move from human browsing heuristics to delegated identity, relationship analysis and continuous seller monitoring.

Key terms

• Agentic Commerce: Agentic commerce is online buying where AI agents discover, evaluate or purchase products on behalf of a person. The governance challenge is that the transaction may be initiated by a non-human actor, which changes how identity, intent, risk and accountability should be interpreted.
• Delegated Intent: Delegated intent is the authority a human gives an AI agent to act within a defined shopping or transaction scope. It is narrower than full autonomy, but it still requires controls that verify what the agent is allowed to do, what it may spend, and how its actions are audited.
• Connected Abuse: Connected abuse is coordinated fraudulent activity that looks harmless when reviewed account by account but reveals a pattern across buyers, sellers, payouts or devices. It often depends on relationship hiding, reused financial details and automation that makes the same actor appear unrelated.
• Marketplace Trust Plane: The marketplace trust plane is the set of catalog, identity, payout and review signals that determine whether a platform can treat a transaction as credible. In agentic commerce, that plane becomes more important because automated discovery and delegated checkout depend on structured, reliable inputs.

What's in the full article

Signifyd's full blog covers the operational detail this post intentionally leaves for the source:

• Step-by-step catalog templating and attribute rules for improving agent discovery across marketplace listings
• Operational guidance for seller verification, payout review and ongoing monitoring of account changes
• Fraud modelling examples showing which human browsing signals lose value in agent-led sessions
• Checkout and post-purchase flow adjustments for delegated transactions and chargeback reduction

👉 Signifyd's full post covers catalog structure, payout controls and fraud signal changes in more implementation detail

Deepen your knowledge

NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance, machine identity security and secrets management. It helps identity and security practitioners design controls for delegated access, lifecycle governance and high-risk identities across modern programmes.