Treat the problem as a cross-channel identity verification issue, not a single-channel fraud problem. Require corroboration across channels, add liveness checks where impersonation risk is high, and make sure approval workflows can resist urgency-based pressure. The strongest controls focus on the decision point, not on employee memory or suspicion alone.
Why cross-channel fraud needs a shared identity verdict
Voice, video and chat are just different surfaces for the same trust decision. If security teams treat each channel independently, they can miss a coordinated impersonation that looks weak in one place but convincing when stitched together. The practical goal is to decide whether the same person, caller or requester is still credible across all channels before any sensitive action is approved.
That means the control point should be the workflow that authorises payment, access, account change or disclosure, not the channel itself. Corroboration works best when one channel confirms details that another cannot easily fake, such as a known callback path, pre-registered contact method, device-bound approval, or a separate trusted verifier.
In practice, this is where Fraud and Cyber Convergence Guide becomes useful: the fraud case and the security case are often the same event, just viewed from different teams. A shared operating model reduces duplicate investigation and helps both groups compare signals before the decision is made.
Where AI-backed impersonation wins, and where it fails
AI-assisted fraud is effective when it can create enough realism to exploit a rushed human decision. A cloned voice may satisfy a quick phone check, a synthetic face may pass a low-friction video meeting, and a polished chat exchange may create confidence through speed and familiarity. The attacker does not need perfect imitation, only enough consistency to push the target past a weak verification step.
It fails when the organisation uses layered verification that is hard to reproduce across channels. Liveness checks help when the interaction must prove that a real, present person is participating, but they are not a silver bullet. Teams should use them where impersonation risk is high, and combine them with verification that is independent of the currently compromised channel.
For teams that need a broader control view, the Collaboration Tool Security Guide is a good reminder that chat, meetings and external access all need explicit governance. The same channel that supports legitimate work can also carry the fraud path, especially when message urgency and meeting pressure are used to defeat judgement.
Strong programmes also map the decision to a known approval boundary. If a request can move money, reset access, change beneficiaries or disclose sensitive data, the workflow should require a second trusted signal outside the live interaction. That is often more reliable than asking staff to “be suspicious” in the moment.
How to make approval workflows resistant to pressure
Urgency-based pressure is one of the most common enablers of this fraud pattern. The attacker creates a sense of immediacy, authority or confidentiality so the reviewer stops validating and starts complying. Once the workflow depends on memory, tone or recognisable style, AI makes the attack easier because it can mimic those cues at scale.
The better approach is to hard-code friction into the decision path for high-risk actions. Require out-of-band confirmation for exceptional requests, separate request initiation from approval where possible, and make high-value actions time-sensitive enough to trigger extra checks rather than bypass them. Where identity confidence drops, the workflow should degrade safely instead of accelerating.
Arup deepfake fraud 2024 is a useful example of why this matters: a persuasive video call can be enough to move a real employee into an irreversible action when the approval design trusts the interaction too much. The lesson is not that video is unreliable, but that video alone is not a sufficient control for consequential approvals.
Where the workflow includes external partners or vendors, treat the contact path as part of the control, not just the person on the line. Callback numbers, approved contact lists, and pre-registered approvers matter because attackers often succeed by inserting a believable alternate route into a legitimate process.
Risk and Threat Considerations
Cross-channel fraud raises the risk of inconsistent verification, especially when one channel is used to reinforce another and employees assume the combined story must be true. AI reduces the cost of creating that consistency, so attackers can pressure a target until a single weak approval step is enough.
Failure mechanism: The control fails when each channel is judged in isolation, the approval path is vulnerable to urgency or authority cues, and no independent trust check exists at the final decision point.
Impact: The result can be payment diversion, account takeover, sensitive disclosure or unauthorised business action, with limited recovery once the action leaves the organisation’s control.
Standards & Framework Alignment
This section maps relevant standards and security frameworks to the operational risks and controls described in this guidance.
OWASP Agentic AI Top 10 addresses the attack and risk surface, while NIST SP 800-53 Rev 5, OWASP ASVS and CIS Controls v8 set the governance and control requirements practitioners need to meet.
| Framework | Control / Reference | Relevance |
|---|---|---|
| OWASP Agentic AI Top 10 | ASI09 — Human-Agent Trust Exploitation | AI-backed fraud exploits human trust across channels to induce unsafe approvals. |
| Recommendation — Add independent verification before any high-impact approval that depends on trusted interaction cues. | ||
| NIST SP 800-53 Rev 5 | IA-5 — Authenticator Management | Cross-channel fraud often succeeds by abusing weak or reusable authentication and recovery paths. |
| AC-2 — Account Management | Fraud can drive unauthorized account changes, resets and privilege updates through approval workflows. | |
| Recommendation — Strengthen authenticator lifecycle controls for high-risk approval and recovery workflows. Restrict who can approve sensitive account changes and require separate verification for exceptions. | ||
| OWASP ASVS | V8 — Authorization | The core failure is an unsafe authorization decision at the point of action. |
| Recommendation — Require robust authorization checks for any action triggered by cross-channel requests. | ||
| CIS Controls v8 | CIS-6 — Access Control Management | Access and approval paths need tighter governance when impersonation is possible across channels. |
| Recommendation — Tighten access approval paths and review exception handling for high-risk changes. | ||
Practitioner Guidance
What to prioritise: Focus on the highest-consequence decision points first, such as payments, payroll changes, beneficiary updates, password resets and privileged access changes. Those are the places where a convincing synthetic interaction is most likely to translate into direct loss.
What to verify: Verify that at least one confirmation step is independent of the live channel being used for the request. If the same chat thread, call or meeting can also carry the approval, the workflow is too easy to manipulate.
Decision rule: If the request is urgent, unusual, confidential or difficult to reverse, require a second verifier and an out-of-band callback before approval. If that step cannot be completed, treat the request as high risk rather than trying to infer intent from the conversation.
What practitioners underestimate: The strongest defence is usually not better human intuition, but a workflow that makes intuition less important. Humans will sometimes be fooled; the process should still prevent a single fooled reviewer from authorising material harm.
Practitioner takeaway: Design fraud controls around the final authority to act, not around the apparent credibility of the voice, face or message in front of the employee.
Related resources from NHI Mgmt Group
- What steps should security teams take to prevent Shadow AI risks?
- How should security teams handle AI-driven identity fraud in remote onboarding?
- How should security teams handle credential sprawl across humans, NHIs, and AI workflows?
- How should security teams handle secret sprawl across cloud and AI workflows?