Online fraud trends for 2026 show AI is reshaping verification

At a glance

What this is: NHI Mgmt Group's analysis, drawing on Veriff's blog post: AI is accelerating impersonation, real-time relay, and injection-style attacks while forcing verification teams beyond static document and MFA checks.

Why it matters: IAM, fraud, and identity verification teams need controls that combine biometric liveness, behavioural analytics, and phishing-resistant authentication because the attack now spans humans, sessions, and injected media.

By the numbers:

• Impersonation fraud accounted for over 85% of all fraud attacks in 2025, according to Veriff.
• Physical AiTM attacks dropped by 34% in 2025, according to Veriff.
• Digital AiTM attacks dropped by 66% in 2025, according to Veriff.
• E-commerce platforms experienced a net fraud rate of 19.2% in 2025, according to Veriff.

Context

Online fraud is shifting from isolated login abuse to layered verification attacks that combine synthetic identity, session interception, and injected media. For identity and access teams, the practical issue is not whether a single control fails, but which part of the verification chain is easiest to manipulate in real time.

Veriff's analysis treats 2026 as a year in which AI changes both sides of fraud detection: attackers gain scale and realism, while defenders gain more signal if they connect document, biometric, device, and behavioural checks. The governance question for IAM and fraud teams is how to make verification adaptive without turning every step into a manual review bottleneck.

For practitioner programmes, the article's value is in showing that traditional assumptions still embedded in verification flows, such as a document image or second factor being enough on its own, no longer hold across the full fraud path.

Key questions

Q: What breaks when identity verification relies on documents or MFA alone?

A: The control fails when attackers can layer AI-generated deepfakes, synthetic identities, session theft, or injected media around a single weak check. Modern fraud does not need to defeat every signal at once. It only needs one blind spot in a fragmented verification flow, which is why teams need correlated evidence across document, biometric, device, and behaviour signals.

Q: Why do AiTM attacks still matter if organisations already use MFA?

A: AiTM attacks matter because MFA can still be bypassed when an attacker relays the user’s authentication flow and steals the resulting session cookie. In that case, the problem is not password strength but session integrity. Phishing-resistant authentication and tighter token controls reduce the chance that a successful login becomes immediate takeover.

Q: How can fraud teams tell whether their verification controls are too easy to spoof?

A: A control is probably too easy to spoof if it trusts one signal in isolation, such as a document image, a face scan, or a device fingerprint, without cross-checking timing and interaction behaviour. Injection and emulator attacks exploit exactly that separation. Strong verification should show disagreement across signals when the input is synthetic or replayed, not silent acceptance.

Q: How should compliance and fraud teams respond when AI-assisted identity fraud increases?

A: They should update risk appetite, escalation paths, and review thresholds together rather than treating fraud as a back-office exception. When AI-enabled deception becomes common, faster approvals must be balanced against loss limits, stronger monitoring, and clearer ownership for suspected synthetic identities.

Technical breakdown

Why impersonation fraud now defeats static identity checks

Impersonation fraud combines stolen or fabricated identity data with counterfeit documents, deepfakes, and synthetic identities. The technical shift is that fraudsters no longer need a single forged artefact to pass a checkpoint; they can assemble a coherent identity narrative across document, face, and behaviour layers. That matters because many verification stacks still evaluate these signals separately, which leaves gaps between detection engines. When the control model is fragmented, attackers tune each artefact to the weakest detector rather than needing to defeat the entire workflow at once.

Practical implication: unify document, biometric, and behavioural signals into one decision path so identity proofing is evaluated as a whole.

How AiTM attacks bypass MFA through session theft

Adversary-in-the-Middle attacks intercept user credentials and session cookies while the victim interacts with a real login flow. Because the legitimate site completes authentication, the attacker inherits a valid session rather than trying to crack MFA directly. That is why phishing kits that relay real-time traffic are so effective against push-based or code-based MFA. The attack is not simply credential theft. It is session theft, which means the defender has to bind the authentication event to the legitimate domain and the legitimate channel, not just to a one-time challenge.

Practical implication: prioritise phishing-resistant MFA and certificate-based authentication where session hijacking is an active threat.

Why emulator and injection attacks change the fraud model

Emulator and injection attacks let adversaries simulate a legitimate device or insert manipulated media into the verification flow. The fraud decision is then made against fabricated telemetry that looks normal unless the platform checks liveness, device authenticity, and interaction timing together. This is a different class of problem from basic account takeover because the attack originates inside the verification session, not before it. In practice, teams need controls that can correlate biometric motion, document handling, and timing patterns across the full interaction rather than trusting any single input stream.

Practical implication: validate device integrity and liveness together, because isolated checks are easy to spoof with injected or replayed inputs.

Threat narrative

Attacker objective: The attacker aims to pass identity verification convincingly enough to obtain authorised access, complete fraudulent transactions, or open accounts under a false identity.

1. Entry occurs when attackers use AI-generated phishing, synthetic identity data, counterfeit documents, or injected mobile environments to reach the verification workflow.
2. Credential or session abuse follows when AiTM kits capture login credentials and session cookies in real time, allowing the attacker to inherit a live authenticated session.
3. Escalation happens when the attacker bypasses the intended verification gates by relaying the victim's interactions or by presenting fabricated biometric and device signals that pass isolated checks.
4. Impact is fraud completion at scale, including account takeover, payment abuse, and successful onboarding or transaction approval in e-commerce and other high-risk channels.

Breaches seen in the wild

• Entra ID actor token flaw (CVE-2025-55241): Hidden Actor tokens plus an Azure AD Graph validation flaw could have let attackers become Global Admin in any Entra ID tenant (CVE-2025-55241).

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

Identity verification is becoming a multi-signal governance problem, not a single-check problem. The article shows that impersonation fraud now blends deepfakes, synthetic identity, and document manipulation, which means no single verification factor carries enough evidential weight on its own. When document, face, device, and behaviour are assessed separately, attackers can tailor each artefact to the weakest gate. For practitioners, verification policy has to be designed around signal correlation, not isolated point controls.

Phishing-resistant authentication is now a fraud-control requirement, not just an IAM hardening measure. AiTM remains effective because the attacker steals the session after the user completes the legitimate login flow, which invalidates assumptions built around traditional MFA. That makes session binding, domain binding, and certificate-backed trust more relevant than one-time challenge mechanisms. Verification programmes that still treat MFA as the end of the control chain are governing the wrong boundary.

Biometric liveness and device authenticity need to be evaluated together to defeat injected media. Emulator and injection attacks succeed when a platform trusts one channel too early, especially on mobile-first flows. The named concept here is verification signal stitching: the practice of combining motion, timing, device, and document evidence into a single fraud decision. Teams that do not stitch those signals create openings for automated manipulation that looks human at the point of decision.

AI cuts both ways, but the governance burden lands on verification architecture. Veriff's article shows fraudsters using AI for realism and scale while defenders use it for anomaly detection and lower false positives. That balance only works when the programme can explain why a decision was made and which signal combination triggered it. For identity leaders, AI does not replace verification design. It raises the bar for evidence quality and decision traceability.

E-commerce fraud rates show that lightweight controls no longer fit high-velocity payment flows. A net fraud rate of 19.2% is not a marginal leakage problem, it is a signal that transaction speed and weak verification are aligned in the attacker's favour. The practical conclusion is that verification and transaction monitoring must be governed together. When payment flows move faster than trust validation, fraud becomes a workflow property rather than an isolated incident.

From our research library:

• Global e-commerce fraud losses are projected to reach $107 billion a year by 2029.

What this signals

Verification signal stitching: The next governance question is not whether a control exists, but whether the controls in the verification chain are stitched together tightly enough to withstand AI-assisted impersonation and session relay. When document, face, device, and session evidence are reviewed as separate events, attackers can route around the weakest one.

Fraud programmes should expect more attacks that look legitimate at the point of decision because the adversary now uses the same tools defenders use, including automation and AI. The practical response is to move verification from a checkpoint mindset to a continuously correlated decision model that can detect injected media, spoofed devices, and relayed sessions in one flow.

For practitioners

• Deploy multi-signal verification for high-risk onboarding Combine facial biometrics, behavioural analytics, and AI-powered document validation so one weak artefact does not decide the entire identity proofing outcome.
• Adopt phishing-resistant authentication for exposed login flows Use certificate-based authentication or other phishing-resistant MFA where attackers can intercept credentials and relay sessions through AiTM kits.
• Test mobile flows against emulator and injection abuse Validate whether device authenticity, liveness, and timing checks still hold when the interaction is replayed, spoofed, or instrumented by automation.
• Tune fraud decisions for fast payment environments Pair real-time transaction monitoring with risk scoring so e-commerce flows can stop abnormal activity before irreversible payment completion.

Key takeaways

• AI has changed online fraud from simple impersonation into a layered verification problem that combines synthetic identities, session theft, and injected media.
• The article's scale signals matter because impersonation dominated 2025 fraud while e-commerce stayed highly exposed, which means weak verification is now a business risk, not a narrow security issue.
• Teams that want to limit fraud need correlated identity proofing, phishing-resistant authentication, and liveness-aware device validation across the whole user journey.

Key terms

• Adversary-in-the-Middle (AiTM): An attack pattern where the adversary sits between the user and the service to capture or replay authentication data and session state. The victim may still log in successfully, which is why the compromise often appears legitimate until privileged actions begin. AiTM is especially dangerous for admin workflows that trust issued sessions too much.
• Verification Signal Stitching: The practice of combining document, biometric, device, timing, and behavioural evidence into one decision model. It matters because modern fraud often bypasses isolated checks by making each individual signal look plausible while the full interaction remains synthetic or manipulated.
• Injected Media: Video, images, or other input data inserted into a verification session to make automation or replay look like genuine user activity. Defenders need to detect this at the interaction layer, not just at the document or face layer, because the injection happens inside the control flow.
• Phishing-Resistant Authentication: Phishing-resistant authentication proves identity without relying on a user to approve a prompt or reveal a reusable secret. It typically binds access to a device, key, or cryptographic proof that an attacker cannot easily reuse or coerce. This approach reduces reliance on human judgment at login time.

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 responsible for identity security strategy or NHI governance in your organisation, it is worth exploring.