Agentic commerce is reshaping checkout, fraud and identity signals

At a glance

What this is: This is an analysis of agentic commerce and its impact on retail fraud, identity signals and checkout control, with the key finding that AI-mediated shopping is changing where trust and abuse now concentrate.

Why it matters: It matters because IAM, fraud and commerce teams now have to govern not just shoppers and sessions, but the identities, intent signals and delegated actions of AI intermediaries across purchase journeys.

By the numbers:

• Sales from those referrals climbed 894% in October 2025.

Context

Agentic commerce is shopping in which an AI agent researches, compares, recommends and sometimes completes a purchase on behalf of a person. The control problem is that the buying journey becomes partially delegated, which weakens the traditional assumptions behind checkout fraud models, attribution and customer verification.

For identity and fraud teams, the issue is not only whether a transaction is legitimate, but whether the merchant can still see who or what initiated the action, how intent was expressed and where responsibility sits when the purchase path is compressed or bypassed. That makes this topic relevant to human identity, delegated access and non-human identity governance alike.

Key questions

Q: What breaks when checkout assumes the buyer is always human?

A: Checkout controls lose context when an AI agent researches, compares and completes purchases on a person's behalf. Behavioural checks, session signals and browsing history become less reliable, so the control model has to shift from human-only interaction to explicit delegation, stronger intent verification and clearer accountability for non-human action.

Q: Why do agent-driven purchases increase fraud and dispute risk?

A: They compress the buying journey and can obscure where intent changed, where an action was authorised and who should be accountable. That makes chargebacks, impersonation and mistaken purchases more likely because the shopper may not recognise the payment context or may not fully understand what the agent approved.

Q: What do retailers get wrong about bot traffic in agentic commerce?

A: They often treat every automated interaction as malicious and miss the difference between authorised delegation and hostile automation. In agentic commerce, some bot activity is legitimate, so the real control challenge is separating approved agent behaviour from takeover, impersonation and abuse.

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

A: They should treat delegated shopping as a joint governance problem. Identity teams define who or what can act, fraud teams score the transaction path, and payments teams manage dispute and authorisation evidence. If those functions operate separately, agentic abuse is harder to detect and harder to explain after the fact.

Technical breakdown

How agentic commerce changes identity and intent signals

Agentic commerce breaks the clean sequence that most ecommerce controls assume. A shopper may set the goal, an agent may browse and narrow choices, and a platform may finish the transaction. That splits identity into at least two layers: the human owner of the intent and the software intermediary carrying it out. In practice, fraud systems that relied on browser behaviour, device consistency or checkout history lose signal quality because the agent can obscure the original context. The core architectural issue is not automation alone, but delegated action without equally strong delegated assurance.

Practical implication: design checkout controls to evaluate both the human source of intent and the non-human actor executing it.

Why site-level fraud controls weaken when purchases are agent-driven

Traditional site-level fraud controls depend on observing the full path through a merchant-owned experience. Agentic flows can bypass that path by comparing products elsewhere, moving through platform intermediaries or completing checkout in a way that never looks like normal browsing. That creates blind spots for velocity rules, high-value thresholds, geo-based checks and behavioural scoring. It also shifts risk upstream into product data quality, platform trust and authorisation design. Where a merchant no longer owns the whole transaction path, fraud control becomes a cross-system governance problem rather than a checkout-only problem.

Practical implication: extend fraud logic beyond the checkout page to the upstream signals that determine how agents are authorised and scored.

Bot-takeover, impersonation and the new abuse surface

The article's abuse scenarios show that agentic commerce introduces a more complicated form of account and identity attack. If fraudsters seize an agentic bot, they can transact at speed under a legitimate-looking account context. If they impersonate agentic purchases through phishing or fake confirmations, they can use trust in the new channel to steal credentials or personal information. That is different from classic carding because the target is not just a payment instrument, but the delegated decision layer itself. The attack surface therefore includes the agent, the communication channel and the customer's expectations of what authorised shopping looks like.

Practical implication: treat agent platforms, purchase notifications and delegated checkout flows as identity-bearing attack surfaces.

NHI Mgmt Group analysis

Delegated shopping is creating a new non-human identity problem. Agentic commerce is not just a faster buying channel. It creates a software intermediary that can act on behalf of a human, which means identity governance now has to account for delegated purchase authority, not only human login events. The practitioner implication is clear: if an action can be initiated by an agent, it needs explicit control boundaries, accountability and revocation logic.

Checkout fraud controls are being pushed upstream into intent verification. In agentic flows, the most important control is no longer only the final authorization decision at payment time. Merchants need confidence in product selection signals, communication integrity and the authorisation conditions that allowed the agent to act. The practitioner implication is to move from checkout-only fraud thinking to journey-wide verification.

Identity signal compression: the loss of intermediate human cues forces merchants to rely on stronger structured data and policy controls. When the agent compresses research, comparison and checkout into a narrow interaction, many of the weak signals fraud teams used to depend on simply disappear. That means the environment rewards cleaner product data, clearer agent permissions and better separation between trusted automation and untrusted automation. The practitioner implication is to govern signal quality as a security control, not just a conversion metric.

Bot abuse now includes trusted automation, not just malicious automation. Older fraud models often treated bots as inherently hostile. Agentic commerce breaks that assumption because some automated activity is legitimate, authorised and user-directed. That complicates abuse detection, but it also creates a sharper governance need: teams must distinguish approved delegation from unauthorised impersonation. The practitioner implication is to classify automation by authority, not by traffic pattern alone.

Retailers need a shared model for identity, fraud and post-purchase governance. The article shows that disputes, returns, identity confusion and site-level fraud evasion can all originate from the same delegated shopping workflow. That makes this a cross-functional control problem spanning IAM, fraud operations, payments and customer support. The practitioner implication is to align policy ownership before agentic commerce scales further.

What this signals

Agentic commerce turns delegated shopping into a governance problem, not just a conversion problem. Retailers will need to decide which agent actions are permitted, which are monitored and which require stronger proof of intent. That makes delegated identity, not only payment risk, the control point that matters most as the channel matures.

What merchants really need is control over trust boundaries. The winning posture will not come from blocking all automation. It will come from distinguishing authorised agents from abusive ones, preserving evidence across the buying journey and making post-purchase resolution easier when humans and software disagree about what was bought.

For practitioners

• Strengthen delegated identity checks Require higher-confidence identity and intent checks when an AI agent initiates research, carting or checkout on behalf of a shopper. Make the delegated action explicit in policy so fraud review can distinguish authorised automation from impersonation.
• Instrument the full buying journey Capture signals before checkout, during comparison and at the point of purchase so you can see where agent-assisted abuse begins. This matters when the purchase no longer runs through one merchant-owned session.
• Update fraud rules for agent behaviour Tune velocity, value and anomaly rules for transactions that arrive through AI-mediated paths instead of standard browser journeys. Preserve review for unfamiliar patterns without treating all agent traffic as hostile.
• Protect against impersonation and fake confirmations Harden customer communications so purchase notices, support links and follow-up messages are verifiable and hard to spoof. That reduces the chance that social engineering turns agentic commerce into a credential theft path.
• Align post-purchase dispute handling Prepare support and risk teams to investigate chargebacks, mistaken purchases and identity confusion arising from agent-assisted orders. Post-purchase clarity becomes part of fraud control when a bot intermediates the transaction.

Key takeaways

• Agentic commerce weakens checkout assumptions because the buyer, the browser and the decision maker are no longer always the same actor.
• The article's risk picture combines visibility loss, bot takeover, impersonation and fraud evasion across the full purchase journey.
• Retailers that want to adapt need stronger product data, clearer identity signals and fraud controls that work before checkout as well as during it.

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.
• Delegated Purchase Authority: Delegated purchase authority is the permission a consumer grants to an assistant, wallet or AI agent to act on their behalf. The security issue is that the delegated actor can become a trusted path for abuse if scope, intent and revocation are not governed tightly.
• Identity signal compression: The reduction of many raw identity events into a smaller set of actionable signals. In practice, this means using analytics to filter noise, cluster related behaviour, and surface the access changes or anomalies that matter most to governance and incident response teams.
• Bot-Takeover: A form of account takeover in which an attacker gains control of an automated agent or bot rather than a normal user session. In commerce, that can let the attacker place purchases, alter preferences or generate fraudulent activity while appearing to operate through a trusted assistant.

Deepen your knowledge

NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance, agentic AI identity and machine identity security. It helps practitioners connect delegated access, identity boundaries and lifecycle control across modern security programmes.