Security teams should combine stronger identity proofing, device and behavior signals, and step-up checks at high-risk moments. AI-driven scams often look legitimate at the surface, so controls need to focus on anomalies across onboarding, payment activity, and account recovery. The goal is to raise friction only where risk is material, while preserving a usable experience for genuine customers.
Why fraud controls need to change when synthetic users can look real
AI-powered scams change the economics of fraud because the attacker can generate convincing messages, identities, and interaction patterns at volume. That means teams can no longer rely on a single surface signal such as a clean login, a plausible profile, or a well-written support request. The control question shifts from “does this look human?” to “does this interaction behave like a trusted customer across multiple moments?”
For fraud operations, the practical challenge is that high-quality deception can pass shallow checks while still revealing itself through cross-channel inconsistency, unusual velocity, or recovery-path abuse. Controls that were tuned mainly for manual review or obvious bot traffic often miss these blended patterns. NIST SP 800-53 Rev. 5 is useful here because it helps teams think in terms of layered access, monitoring, and response controls rather than a single gate. In practice, many security teams encounter the weakness only after synthetic activity has already blended into normal customer journeys.
How fraud teams should re-balance signals and interventions
The most effective adaptation is to treat fraud prevention as a decisioning problem, not a single authentication problem. AI-driven scams can imitate language and timing, but they still struggle to maintain consistency across identity proofing, device history, network reputation, behavioral entropy, and transaction context. Teams should therefore combine signals that are difficult to fake together, and reserve stronger friction for moments where the business impact of abuse is highest.
This usually means tightening controls around onboarding, account recovery, payment changes, payout requests, and credential reset flows. Those are the points where a synthetic actor can convert initial trust into real loss. The key design choice is to avoid blanket friction that punishes genuine users. Instead, teams should use risk-based escalation: low-friction verification for normal activity, and additional proof when the transaction or identity trail deviates from established patterns.
- Use identity proofing to separate first-time trust from repeat trust.
- Correlate device, session, and behavior history before allowing high-value actions.
- Apply step-up checks only where the abuse case justifies the added friction.
- Feed review outcomes back into the scoring model so the control adapts over time.
This approach works best when the fraud stack is connected to customer lifecycle events rather than only perimeter or login events. It breaks down when teams score isolated events without enough context to distinguish a real customer in a new situation from an automated scam that has learned to imitate ordinary behavior.
Where AI-driven fraud changes the control trade-off
Tighter fraud controls often increase friction and review load, so organisations have to balance loss prevention against conversion, support burden, and false positives. That trade-off becomes sharper when scams are operating at scale, because the same control that blocks abuse may also interrupt legitimate customers during recovery or payment. The right answer is not always “more checks”; it is “better-timed checks.”
There is still some disagreement in the industry about how much weight to give biometric-like behavioral signals versus more traditional identity evidence. The consensus is stronger on one point: no single signal is reliable enough on its own when the adversary can generate many plausible variants. A useful external distinction is between signals that prove continuity of a customer relationship and signals that only show a momentary interaction that appears normal.
Another edge case is delegated or shared access. A legitimate user may look unusual because they are acting through a family device, a corporate network, or an accessibility workflow, while an attacker may look normal by reusing common tools and patterns. Teams need exception handling for these cases, or else the control will either miss abuse or over-block genuine users.
Practitioners should expect the control to degrade when review thresholds are static, when recovery channels are weakly governed, or when fraud telemetry is not fed back into identity and access decisions.
Risk and Threat Considerations
AI-powered scams create two linked risks: identity impersonation at scale and the erosion of trust in customer-facing workflows. The material exposure is not just account takeover, but also false legitimacy in onboarding, support, payments, and recovery paths. As synthetic content improves, the main danger is that organisations overtrust surface similarity and underweight contextual inconsistency.
Failure mechanism: Adversaries use generated text, reused personal data, scripted interactions, and automation to satisfy shallow checks while progressively building credibility across multiple touches. They then exploit weak recovery flows, high-trust exceptions, or delayed human review to convert that apparent legitimacy into fraud.
Impact: Organisations can lose funds, approve fraudulent accounts, expose customer data, and degrade the reliability of fraud scoring itself. Once the control set starts accepting synthetic interactions as normal, it becomes harder to separate genuine users from coordinated abuse.
Standards & Framework Alignment
This section maps relevant standards and security frameworks to the operational risks and controls described in this guidance.
MITRE ATT&CK and OWASP Non-Human Identity Top 10 address the attack and risk surface, while NIST CSF 2.0 and CIS Controls v8 set the governance and control requirements practitioners need to meet.
| Framework | Control / Reference | Relevance |
|---|---|---|
| NIST CSF 2.0 | PR.AA — Identity Management, Authentication, and Access Control | Fraud controls depend on trustworthy identity and access decisions. |
| DE.CM — Continuous Monitoring | AI scams are best detected through correlated anomalies over time. | |
| RS.MI — Mitigation | Fraud workflows need rapid containment when abuse is confirmed. | |
| Recommendation — Strengthen identity and access checks where customer trust can be converted into loss. Correlate session, device, and transaction telemetry to detect synthetic behavior. Trigger step-up friction and containment when abuse patterns cross a material threshold. | ||
| CIS Controls v8 | 6 — Access Control Management | Fraud often exploits weak recovery and step-up access paths. |
| 8 — Audit Log Management | Detection depends on preserving evidence across fraudulent journeys. | |
| Recommendation — Restrict sensitive customer actions to verified, risk-appropriate access paths. Log identity, device, and transaction events well enough to reconstruct abuse chains. | ||
| MITRE ATT&CK | T1036 — Masquerading | Scams mimic legitimate users and activity to blend in. |
| T1110 — Brute Force | Scaled scams often use automation to test credentials and recovery paths. | |
| Recommendation — Map impersonation patterns to masquerading behavior and hunt for lookalike activity. Detect repeated access attempts against login and recovery workflows. | ||
| OWASP Non-Human Identity Top 10 | NHI-01 — Secrets and Credential Management | Fraud campaigns often abuse tokens, OTPs, or recovery credentials as trust shortcuts. |
| Recommendation — Treat recovery tokens and similar secrets as high-value trust assets and protect them tightly. | ||
Practitioner Guidance
What to prioritise: Focus first on the workflows where trust is converted into loss, especially onboarding, recovery, payout, and payment-change events. Those are the places where AI-driven scams most often bypass generic front-door checks and where incremental friction has the highest defensive value.
What to verify: Test whether your strongest signals are actually independent. If device, identity, and behavior checks all rely on the same weak source, the control is more fragile than it appears. Validate that step-up decisions are based on combinations of evidence, not a single “suspicious” flag.
What good looks like: Strong programs show lower false confidence in clean-looking sessions and higher confidence in context-rich decisions. They can explain why a case was escalated, preserve customer usability for normal activity, and update thresholds as attack patterns change.
Practitioner takeaway: The best fraud controls do not try to detect “AI” directly; they make it hard for any synthetic actor to sustain trust across the full customer journey.
Related resources from NHI Mgmt Group
- How should fraud teams adapt controls when AI-powered attacks scale faster than review capacity?
- How should security teams stop agentic AI fraud without blocking real users?
- How should security teams reduce AI-powered fraud in SaaS applications?
- How should security teams protect browser-side fraud controls against AI analysis?
Deepen Your Knowledge
Reviewed and updated by the NHIMG editorial team on September 7, 2026.
NHI Mgmt Group — the #1 independent authority on Non-Human Identity, IAM, and Agentic AI security. nhimg.org
