Ecommerce is splitting into human and AI agent buyers in 2026

At a glance

What this is: Signifyd’s 2026 commerce analysis says online retail is moving toward a mixed buyer environment where humans and AI agents both shape discovery, checkout and fraud risk.

Why it matters: This matters to identity and fraud teams because trust signals built for human sessions are weaker when agents browse, compare and act on behalf of customers.

By the numbers:

• Same-store ecommerce spending grew 6% last year, even with inflation sitting at 2.7% by December.
• The top 10% of earners now account for almost 50% of all consumer spending in the U.S., up from about 33% in the early 1990s.
• ATO fraud grew 45% year over year in apparel, 31% in grocery and 31% in electronics in 2025.

Context

Ecommerce is no longer governed by a single buyer model. Human shoppers still matter, but AI agents are increasingly entering the purchase path, which changes how merchants should think about trust, intent and fraud detection.

The identity problem here is not just who signs in, but who is acting for whom. That creates a governance gap between session-based controls built for people and transaction controls that must also assess delegated automation.

For fraud and IAM teams, the practical issue is that agent-driven journeys often produce thinner behavioural context, so detection and authorisation logic has to shift earlier in the interaction.

Key questions

Q: What breaks when fraud controls are built only for human browsing sessions?

A: Fraud controls built only for human browsing sessions lose critical context when AI agents act on behalf of customers. Device fingerprints, cookies, and IP reputation may disappear or become unreliable, which makes legitimate delegated activity look suspicious and can also hide malicious automation. Teams need identity-aware delegation signals, not just browser-era telemetry.

Q: Why do AI agents create more risk around trust and checkout decisions?

A: AI agents compress the shopping journey and often reduce the behavioural evidence merchants rely on to validate intent. That increases the chance of both false declines and missed fraud, especially when product data and policy metadata are inconsistent. The risk is not the presence of automation itself, but the mismatch between automated buying and controls designed for human browsing and manual checkout.

Q: How do you know if account takeover controls are actually working?

A: Look for reduced successful takeovers, lower fraud losses, and preserved good-user throughput at the same time. If false positives rise sharply or attackers simply shift tactics while account compromise stays flat, the control is creating friction without changing outcomes.

Q: How do security teams make risk scores actionable for both people and AI agents?

A: Risk scores become actionable when they are tied to plain-language explanations and a clear response path. Teams need to know why a score is elevated, what changed, and what action should follow. That makes interventions easier to defend, helps owners understand the issue, and turns scoring into a working governance control.

Technical breakdown

Why AI agent traffic changes ecommerce trust models

AI agents alter the trust model because they collapse discovery, comparison and sometimes checkout into a machine-mediated path. Traditional ecommerce controls infer legitimacy from browser behaviour, click patterns and session depth, but those signals are weaker or absent when an agent is acting on behalf of a person. Merchants therefore need to evaluate intent, provenance and transaction context rather than relying on human browsing cues alone. In practice, this shifts trust from the session to the request, and from page behaviour to structured product and order signals.

Practical implication: move fraud and trust evaluation upstream to the request layer instead of depending on human-style session heuristics.

How agentic commerce exposes gaps in product data and policy logic

Agentic commerce depends on machine-readable inputs. Clean titles, consistent attributes, accurate pricing, clear shipping promises and readable return terms all help an agent decide whether to recommend or purchase a product. If those fields are inconsistent, the merchant becomes less discoverable to agents and less trustworthy to automated decisioning. The same data quality issues also complicate fraud screening, because policy logic cannot distinguish a legitimate agent-led purchase from an anomalous automated attempt if catalog and checkout signals are noisy.

Practical implication: treat product data quality and policy consistency as part of the fraud control stack, not just merchandising hygiene.

Why account integrity is now a lifecycle problem

Account takeover in ecommerce works because dormant or weakly monitored accounts create reusable trust. Once criminals gain access, they can exploit stored payment instruments, loyalty balances and gift cards without needing to defeat checkout controls. That makes identity lifecycle governance central to fraud defence. Monitoring new-device logins, unusual geolocation, velocity changes and profile edits gives merchants a better chance of stopping abuse before the account is monetised. The control boundary has moved from checkout to the whole account journey.

Practical implication: extend identity monitoring beyond login to include behavioural and profile-change signals across the full account lifecycle.

NHI Mgmt Group analysis

AI agent shopping creates a verification trust gap: ecommerce fraud controls were built to interpret human behaviour, but agent-led journeys remove many of those cues. That does not make the journey untrustworthy by default, but it does make legacy session scoring less reliable. Practitioners should treat delegated shopping as a distinct trust condition with its own controls.

Account integrity has become a governance issue, not just a fraud issue: when dormant accounts, stored credentials and loyalty balances can be monetised at scale, identity lifecycle discipline becomes part of revenue protection. The merchant no longer manages only authentication events, but the persistence of trust across long-lived customer accounts. Teams should align fraud operations with identity governance.

Product data quality is now an authorisation-adjacent control for commerce automation: agents can only act on what they can interpret, so structured attributes and consistent policy metadata influence whether automated buyers can safely transact. Messy catalog data creates both conversion loss and control ambiguity. Practitioners should manage catalog integrity as a trust input, not a merchandising afterthought.

Autonomous checkout will reward merchants that can prove intent, not just identity: as checkout becomes more automated, the decisive question shifts from who is present to what authority the actor has to buy. That is a familiar identity problem in a new form. Teams should prepare for transaction governance that separates delegated automation from abuse.

Human and AI buyers will fragment fraud strategy: one control set will not serve both populations well. Human buyers need customer experience-aware risk controls, while AI agents need structured trust rules and clearer machine-readable policies. The market will increasingly split into merchants that adapt their governance to both and merchants that keep tuning for only one. Practitioners should plan for dual-path commerce risk management.

What this signals

AI-agent commerce will force a control split. Merchants can no longer assume that the same behavioural signals will support both human and machine-mediated shopping. The practical response is to design trust decisions around delegated intent, then tune fraud rules and account protections to that distinction.

Dormant accounts are the new fraud reservoir. As more commerce shifts toward long-lived customer profiles, old accounts with stored value become the easiest monetisation path for attackers. Teams should expect the highest return on detection work to come from upstream account monitoring rather than checkout-only controls.

For practitioners

• Separate human and agent trust paths Define different trust and risk decisioning for human shoppers and AI-agent-driven sessions, especially where behavioural signals are thinner or absent.
• Harden account lifecycle monitoring Track dormant accounts, new-device logins, email and phone changes, shipping edits and unusual redemption behaviour as part of account takeover defence.
• Improve machine-readable product data Normalize product titles, attributes, pricing, shipping promises and return terms so agentic buyers can interpret offerings without ambiguity.
• Use risk-based returns decisioning Route low-risk return requests for fast resolution, push ambiguous cases into review and decline clear abuse patterns instead of applying one policy to all cases.

Key takeaways

• Ecommerce in 2026 is becoming a dual-audience environment, with human shoppers and AI agents driving different trust and fraud requirements.
• The most exposed merchant control points are account lifecycle events, machine-readable product data and transaction decisioning that depends on human-style session signals.
• Merchants that separate human and agent paths will be better positioned to reduce false declines, protect accounts and limit abuse without slowing legitimate commerce.

Key terms

• Agentic Commerce: Agentic commerce is a buying and transaction model where software agents act on behalf of a person. The identity challenge is not just proving who owns the account, but constraining what the agent may do, for how long, and under what revocation and audit rules.
• Account Takeover: Account takeover is unauthorized use of a legitimate account after an attacker obtains valid access through stolen credentials, tokens, or trusted integrations. The key security problem is that the resulting activity often looks normal to logs and controls, which makes containment and attribution harder than in a forced-entry breach.
• Delegated Intent: Delegated intent is the machine-verifiable statement of what a human authorised an agent to do, for which purpose, and within which scope. It becomes a governance boundary when linked to the principal, the resource, the trust level, and the policy version active at the time of execution.

Deepen your knowledge

NHI Mgmt Group covers identity security, NHI governance, and agentic AI through independent research, practitioner guides, and the NHI Foundation Level course. Explore nhimg.org for resources that connect identity governance to the broader security disciplines your programme depends on.