AI agents and bad bots are widening retail fraud exposure

At a glance

What this is: This is Arkose Labs' analysis of how bad bots and AI agents are expanding retail fraud exposure through higher login abuse, fake account creation and automated payment attacks.

Why it matters: IAM, fraud and digital identity teams need to treat machine-driven abuse as a first-class access problem because account protection, challenge strategy and signal sharing now sit inside the same control plane.

By the numbers:

• 65% of e-commerce traffic comes from bad bots.
• Login attack rates at e-commerce sites jumped 216% in 2025.
• Annual e-commerce losses to online payment fraud reached $48B.
• Arkose Labs uses 250+ signals to identify suspicious activities in real time.

Context

Retail fraud is no longer just a human attacker problem. When bad bots generate most of the traffic, login abuse and fake account creation become part of the access layer, not a separate threat category.

That changes the governance question for e-commerce and digital identity teams. Controls built only around passwords, MFA prompts or static bot filters do not fully address how machine-driven abuse now shapes account compromise, transaction fraud and customer trust.

Key questions

Q: What breaks when bad bots dominate e-commerce traffic?

A: Fraud controls lose signal quality when machine traffic overwhelms normal customer behaviour. Login, registration and checkout decisions become less reliable, and the organisation starts measuring volume instead of intent. That means account takeover, fake accounts and payment abuse can scale before the platform recognises the pattern.

Q: Why do automated retail fraud attacks create more risk than traditional loss prevention tactics can handle?

A: Automated fraud increases risk because it raises attack speed, scale, and sophistication faster than manual or store-centric controls can respond. Techniques built for brick-and-mortar loss prevention often fail online because digital attacks can be repeated quickly, adapted in real time, and distributed across channels. That makes prevention dependent on continuous scoring and faster decisioning.

Q: What are the signs that bot mitigation is failing in a live environment?

A: Bot mitigation is failing when malicious traffic looks normal enough to evade basic filters, suspicious activity keeps recurring, and teams only see the problem after impact. Another warning sign is when defenders rely mainly on reactive blocking instead of proactive alerting. If legitimate users are disrupted while attackers keep adapting, the control is not keeping pace.

Q: What do teams get wrong about balancing fraud prevention with customer experience?

A: A common mistake is overcorrecting with controls that stop fraud but also block legitimate buyers. That approach can protect a narrow risk metric while damaging conversion, revenue, and trust. Better programs tune controls to the actual fraud profile, use risk-based decisions where possible, and continuously test whether the experience remains streamlined and secure.

How it works in practice

Why bad bot traffic distorts identity trust signals

Bad bots do more than create volume. They distort behavioural baselines by mixing automated browsing, credential stuffing, fake registration flows and scripted purchase activity into traffic that can resemble legitimate customer journeys. In practice, this makes signal quality the first control problem: if the platform cannot separate human intent from machine activity, every downstream decision becomes less reliable. The article's references to 250+ signals and real-time detection point to a layered scoring model rather than a single rule. Practical implication: decisioning has to weight identity risk before account creation, login and checkout, not after abuse has already propagated.

Practical implication: shift fraud controls toward pre-authentication and pre-transaction signal analysis.

How adaptive challenges change the economics of fraud automation

Adaptive challenges are designed to increase attacker cost without creating unnecessary friction for legitimate users. They work by adjusting the verification burden based on observed behaviour, device traits and session context, so scripted abuse becomes more expensive to scale. That matters because bot operators optimise for throughput and low unit cost, while retailers need to preserve conversion and customer experience. The article's emphasis on stopping fraudulent transactions and fake account creation shows the control point is economic as much as technical. Practical implication: tune challenge intensity to specific abuse patterns, not to a single global threshold.

Practical implication: calibrate step-up friction by abuse pattern, channel and risk score.

Why fraud prevention now overlaps with account takeover and API security

Retail fraud often traverses multiple surfaces in one chain. A bot may start with credential stuffing, move into account takeover, then abuse APIs or checkout flows to monetise access. That is why account takeover, fake account creation, SMS toll fraud and API security belong in the same operating model rather than separate teams with disconnected telemetry. The article makes that convergence explicit by bundling these outcomes under one platform narrative. Practical implication: correlate identity, device and API telemetry so one abuse pattern cannot hide behind siloed ownership.

Practical implication: unify account, API and device telemetry into a single fraud decision pipeline.

NHI Mgmt Group analysis

Retail fraud is now an identity governance problem, not just a bot management problem. When bad bots account for most of the traffic and login attacks are rising sharply, the practical boundary has moved from perimeter filtering to trust governance. E-commerce teams must treat account creation, login and checkout as one identity journey with one fraud decision model. The implication is clear: fraud, IAM and application security can no longer operate as separate control planes.

Machine-driven abuse collapses the old assumption that customer risk is primarily human-paced. Controls were built for discrete user events, but bot activity can test, adapt and iterate faster than review cycles or manual escalation paths. That makes static thresholds a weak answer because attackers can tune around them. Practitioners should read this as a signal that velocity, not just volume, is now the defining property of retail identity abuse.

Adaptive friction is becoming the economic control point for online fraud. The most useful control is not perfect blocking, but raising attacker cost enough that abuse is no longer profitable. Arkose Labs' signal-driven approach reflects a broader market shift toward risk-based interaction control, where challenge intensity is tuned by context rather than applied uniformly. For practitioners, the key question is whether their current stack can price out automation without breaking the customer journey.

Identity blast radius is the right named concept for this category of exposure. Once automated abuse reaches login, account creation or API-backed checkout, the damage is no longer limited to one session or one endpoint. A weak decision model lets a cheap bot swarm spread into account takeover, fake accounts and transaction fraud across the business. The implication is that governance must contain abuse at the first trust decision, not after monetisation has begun.

From our research library:

• 53% of security leaders expect AI to run major portions of their infrastructure autonomously within the next three years, according to the 2026 Infrastructure Identity Survey.
• 69% of security leaders agree identity management must fundamentally shift to address agentic AI systems, according to the 2026 Infrastructure Identity Survey.
• Read next: Agentic AI Security Guide

What this signals

Identity and fraud teams should assume automation is already inside the customer journey. When bot traffic dominates and login abuse is accelerating, the practical control question is not whether to block automation, but where to force a higher-confidence trust decision. That shifts emphasis toward device reputation, behavioural scoring and adaptive challenge orchestration, with API and checkout telemetry feeding the same pipeline.

Identity blast radius: once machine traffic reaches login, account creation or payment, one cheap abuse path can cascade into account takeover, fake accounts and monetised fraud. Teams should map which customer journeys still rely on static thresholds, because those are the paths attackers will keep industrialising.

Retail programmes that separate fraud operations from IAM governance will keep missing the same pattern. Access decisions at login, registration and checkout now determine whether automation can convert into loss, so the control boundary has moved upstream.

For practitioners

• Instrument pre-authentication risk scoring Score traffic before login and registration so bot-heavy sessions can be challenged before they reach account creation or payment flows.
• Separate customer and automation signals Combine device, behavioural and session features so scripted traffic is not evaluated with the same thresholds as trusted customer activity.
• Correlate account and checkout telemetry Tie login abuse, fake account creation and payment events into one investigation path so a single abuse campaign cannot be treated as isolated incidents.
• Tune step-up controls to fraud economics Use adaptive challenges where the cost to the attacker matters most, especially for high-value accounts, promotions, and payment flows.

Key takeaways

• Bad bots and AI-driven automation are no longer background noise in retail. They now shape the trust conditions under which fraud, account takeover and payment abuse succeed.
• The article ties that exposure to concrete scale indicators, including 65% bad-bot traffic, a 216% jump in login attack rates and $48B in annual e-commerce fraud losses.
• Retail teams need one decision model across identity, device and transaction layers, or attackers will keep turning low-cost automation into high-cost abuse.

Key terms

• Bad Bot: An automated client designed to imitate normal user behaviour while carrying out abuse at scale. In identity and fraud programmes, bad bots distort telemetry, inflate traffic, and can drive credential stuffing, fake account creation, and checkout abuse. Their impact is operational and governance related, not just volumetric.
• 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.
• Adaptive Challenge: A response mechanism that increases friction only when a session or request shows elevated risk. It relies on telemetry, behavioural scoring, and context so that legitimate users are not unnecessarily blocked while automation faces stronger verification or challenge steps.
• Identity Blast Radius: The amount of damage a compromised identity can cause across systems, data, and infrastructure. In NHI environments, it is shaped by permissions, network reach, and administrative capability rather than by the credential alone. Reducing blast radius is a containment strategy that limits lateral movement and data exposure.

Deepen your knowledge

NHI governance, agentic AI identity, and machine identity lifecycle are core topics in our NHI Foundation Level course, the industry's only accredited NHI security programme. If you are building or maturing an IAM programme, it is worth exploring.