AI-enabled retail fraud is scaling past legacy checks

At a glance

What this is: This is a retail fraud analysis showing how AI-powered attacks are automating return fraud and other scams to look like ordinary customer activity while scaling volume fast.

Why it matters: It matters because identity and fraud teams must treat bot-driven fraud as an access and verification problem, not only a transaction review problem, especially where customer trust signals are easy to imitate.

By the numbers:

• Among Pindrop customers in retail, there was a 56% month-over-month increase in non-live fraud in November.
• Live fraud dropped by 69% in the same period among Pindrop retail customers.

Context

Retail fraud now blends automation, impersonation and low-friction account abuse, which makes it harder for legacy controls to distinguish genuine customers from scripted activity. Traditional checks were built for slower, more obviously human attack patterns, not for high-volume fraud that can adapt to thresholds, workflows and review queues. Online fraud attacks is a useful lens here because the problem spans return fraud, account abuse and impersonation.

In this article, the core governance gap is not just fraud detection accuracy but identity assurance under adversarial automation. When attackers can behave like normal customers at scale, controls must evaluate intent, velocity and behavioural consistency, not only credentials or one-time verification events. The retail starting point is increasingly typical, not exceptional.

Key questions

Q: What breaks when retail fraud becomes automated and AI-assisted?

A: Manual review, threshold-based approval and one-time verification all become easier to game when attackers can vary requests and repeat them at machine speed. The main failure is not a single control collapse but the accumulation of many low-risk-looking events that individually appear normal. Defenders need workflow-level telemetry and fraud rules that account for repetition, not just authenticity at a point in time.

Q: Why do passwords, KBAs and OTPs struggle against synthetic customer behaviour?

A: Those controls answer a narrow question about proof at a single moment. They do not reliably detect whether a requester is an AI-assisted bot, a deepfake operator or a repeated attacker reusing the same workflow. When fraudsters can imitate normal behaviour, assurance has to include behavioural context, device consistency and transaction pattern analysis.

Q: How can teams tell whether AI-assisted fraud is becoming a practical problem?

A: Look for faster campaign iteration, more personalised impersonation, and higher retry volume across email, chat, and support channels. Those signals suggest attackers are using AI to improve effectiveness. If the same campaigns also lead to credential resets, payment attempts, or privilege requests, the problem has moved from content to identity abuse.

Q: Should organisations treat deepfake fraud as a fraud team issue or an identity governance issue?

A: Both, but the governance owner needs to be identity because the control failure happens at verification time. Fraud teams can tune detection, yet identity leaders must define the assurance threshold, the escalation path, and the evidence required before trust is granted.

Technical breakdown

How AI-driven return fraud works at scale

AI-driven return fraud uses automation to submit repeated refund or return requests, often through bots that imitate ordinary browsing and purchase patterns. The attacker does not need to defeat every control in one move. Instead, they optimise for low-value, high-volume abuse that stays beneath review thresholds and blends into expected customer behaviour. Machine learning can help tune timing, content variation and request sequencing so the fraud looks less synthetic. That makes the attack economically attractive even when each individual transaction seems trivial. The control problem is therefore pattern recognition across many weak signals, not single-event authentication alone. Read Top 10 Online Fraud Attacks.

Practical implication: Prioritise behavioural analytics and velocity-based controls for return workflows rather than relying on one-time verification at the front door.

Why legacy checks fail against synthetic customer behaviour

Passwords, knowledge-based authentication and one-time passcodes all assume the verifier can reliably tell a real customer from an impostor at a discrete decision point. AI-enabled fraud weakens that assumption because the attacker can present correct credentials, generate convincing dialogue, and repeat attempts at scale until one path succeeds. These controls may still have value, but they are poor at detecting coordinated abuse across accounts, devices and sessions. In fraud operations, the question is not just whether an identity can authenticate. It is whether the entire interaction behaves like a legitimate customer journey over time.

Practical implication: Treat passwords, KBAs and OTPs as supporting controls, not as sufficient fraud barriers for high-volume retail return flows.

What non-live fraud signals reveal about adversarial automation

A shift from live fraud to non-live fraud suggests attackers are optimising for channels where they can automate repeatable steps and avoid immediate human challenge. Non-live activity often includes scripted submissions, fabricated support interactions and other interactions that can be replayed or varied cheaply. When that trend rises, it usually means the attacker has found a workflow that is easy to scale and difficult to detect with manual review alone. The important technical point is that fraud now behaves like a distributed campaign, not a single transaction anomaly. Detection has to move upstream into device reputation, workflow integrity and friction tuning.

Practical implication: Use the fraud mix shift as a signal to tighten review thresholds, device intelligence and workflow controls before losses compound.

Threat narrative

Attacker objective: To extract repeated low-dollar refunds and other retail value at scale while staying below suspicion thresholds.

1. Entry occurs through retail return or customer-service workflows where bots can submit requests that resemble ordinary consumer behaviour.
2. Credential and identity checks are bypassed or satisfied with synthetic, AI-generated interactions that reduce the chance of immediate challenge.
3. The attacker escalates by repeating low-value refund attempts at volume, turning each small success into a scalable abuse stream.
4. Impact appears as compounded financial loss, operational noise, and degraded trust in existing fraud controls.

Breaches seen in the wild

• Arup deepfake fraud 2024: Deepfakes of Arup's CFO and colleagues on a video call led a Hong Kong employee to transfer HK$200 million (about US$25.6m) to fraudsters.

Read and download The State of NHI & AI Agent Breach Report 2026, covering 200+ breaches impacting Non-Human Identities including AI Agents.

NHI Mgmt Group analysis

AI fraud is now a workflow problem, not just a verification problem. The article shows attackers using bots and AI to imitate ordinary customer behaviour, which means the control failure sits inside retail processes rather than only at login. That shifts the governance question from “did the customer authenticate” to “did the transaction behave like a genuine customer journey.” Identity teams should treat behavioural consistency as part of assurance, not a separate fraud layer.

Return fraud is a classic example of compounding loss economics. A single refund may be too small to trigger review, but AI lets attackers repeat the same pattern until the total becomes material. That creates a governance blind spot where thresholds designed to reduce friction also create scale for abuse. Practitioners should recognise that loss aggregation is itself the attack method, not just a side effect.

Deepfake and scripted fraud now blur the line between human and machine impersonation. The article’s reference to deepfake calls and non-live fraud shows that identity assurance must account for synthetic behaviour as well as stolen credentials. Deepfakes, social engineering and AI impersonation becomes relevant where voice, video and conversational cues are part of the trust decision. Retail programmes need to assume that fraudsters can manufacture believable interaction, not only fake identity attributes.

Retail fraud telemetry should be read as an identity signal, not only a revenue signal. A 56% increase in non-live fraud alongside a 69% drop in live fraud suggests attackers are rebalancing toward automation-friendly channels. That pattern should drive controls around device fingerprinting, session behaviour and workflow integrity, with clearer ownership between fraud, IAM and customer operations.

From our research library:

• Nearly 60% of companies reported that fraud losses were still increasing in 2025.

What this signals

Retail fraud programmes now need a stronger trust boundary around customer workflows. AI makes the old split between authentication and fraud detection too narrow, because the attacker can satisfy both while still abusing the process. The programme question is not whether a customer can prove who they are once, but whether the full interaction remains trustworthy across repeated steps.

Compounding-loss fraud is the new operational risk pattern. Small, individually acceptable refunds can become material losses when bots can sustain the attack continuously. That means control owners should think in terms of cumulative exposure, not only transaction-level exception handling.

For practitioners

• Tighten return-workflow controls Add velocity limits, anomaly scoring and step-up review on refund and return flows that can be scripted at scale.
• Instrument behavioural fraud signals Correlate device reputation, request timing, session reuse and customer journey deviations to spot AI-generated abuse earlier.
• Revisit low-friction verification points Review where passwords, KBAs and OTPs are being treated as sufficient proof in high-risk retail journeys.
• Strengthen deepfake-resistant verification Use out-of-band checks and payment approval rules where voice or video interaction can influence refunds or account changes.

Key takeaways

• AI-driven retail fraud now scales through normal-looking workflows, which makes behavioural verification as important as credential checks.
• The strongest evidence in the article is the shift toward non-live fraud and away from live fraud, which indicates automation is changing the attack mix.
• Retailers should respond by tightening return controls, improving behavioural telemetry and using stronger verification where synthetic impersonation is plausible.

Related NHI Mgmt Group resources: Top 10 Online Fraud Attacks · Deepfakes, Social Engineering and AI Impersonation Guide · Human vs Non-Human Identity

Key terms

• AI-Enabled Fraud: AI-enabled fraud is deception carried out with the help of artificial intelligence to make scams faster, more convincing, or harder to detect. It includes synthetic identities, deepfakes, automated phishing, and adaptive social engineering. In security terms, it combines machine-generated content with identity abuse, fraud operations, and evasion of human and technical controls.
• Non-live Fraud: Fraud carried out without a live human interaction at the point of execution, often through bots, scripted requests or automated tooling. It matters because these attacks can be repeated at volume, varied cheaply and hidden inside flows that were designed for ordinary customer activity.
• Return Fraud: Return fraud is the abuse of merchant return policies to obtain refunds, replacements, or store credit without a legitimate basis. It can include exaggerated damage claims, item substitution, wardrobing, and AI-assisted narratives that make a weak claim appear credible to review teams.
• Deepfake-based impersonation: A fraud technique that uses synthetic audio, video, or both to make an attacker appear to be a trusted person during a live interaction. The tactic exploits human trust in familiar cues and often aims to trigger urgent actions such as payments, resets, or access changes before verification is challenged.

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. It is suitable for practitioners who need to connect identity assurance with the broader security programme.