Deepfake fraud exposed the limits of face-based identity proof

At a glance

What this is: Idemia examines the Arup deepfake fraud case and shows how video, voice and ordinary approval checks can still leave identity unverified.

Why it matters: IAM and finance teams should treat live video as an input to verification, not proof of identity, and add independent checks for high-risk approvals across human and delegated workflows.

By the numbers:

• A finance employee authorized 15 wire transfers totaling HK$200 million, about US$25.6 million, after the deepfake video call.

Context

The core governance gap here is simple: familiar appearance is not identity proof. In this case, attackers used public audio and video to impersonate senior executives, then paired that impersonation with urgency and a video call to bypass human judgment in a high-value payment workflow.

For identity and access teams, the relevance goes beyond fraud. It shows that workflows built around recognition, approval chains and human confidence can still fail when the requestor is not independently authenticated, even if the interaction looks normal in real time.

That makes this a human identity and transaction assurance problem, not a perimeter breach. The question for practitioners is how to separate communication from verification when the business process itself depends on trust signals that can now be synthesized convincingly.

Key questions

Q: What breaks when a deepfake video call is used to authorize a payment?

A: The approval chain breaks because the request appears to come from a trusted person even though the identity behind it has not been independently verified. That means transaction controls can pass while authenticity fails. Organisations should require proof that is separate from the call itself before releasing money or changing critical records.

Q: Why are familiar faces and voices no longer enough for transaction approval?

A: Because synthetic media can recreate those cues convincingly enough to trigger trust without proving presence. For high-risk approvals, the business risk is not just impersonation, but the collapse of the assumption that human recognition is a reliable identity signal. Teams need evidence that is independent of the live interaction.

Q: What are the signs that identity verification is too weak for a growing digital business?

A: Common warning signs include repeated fraudulent account openings, account takeovers, inconsistent approval decisions across markets, and heavy manual review that slows onboarding. If verification cannot keep pace with transaction growth, organisations often see rising fraud losses or customer drop-off. A weak programme also struggles to support expansion into new geographies, where compliance and identity signals vary significantly.

Q: How should organisations verify executive requests before releasing funds?

A: Use an out-of-band confirmation path that is pre-agreed, routinely tested and separate from the conversation that initiated the request. A second approver, trusted callback number or cryptographic proof tied to the requester gives the business a stronger basis for release than recognition alone.

Technical breakdown

How deepfake fraud defeats human recognition

Deepfake fraud works by substituting synthetic audiovisual evidence for the social cues people normally use to judge legitimacy. Attackers first gather public recordings, then model a target’s face and voice, and finally use that synthetic presence in a live interaction to create the illusion of continuity. The important detail is that the call can be interactive and still be fake. That means real-time conversation no longer guarantees the speaker is present, unique, or authorised, which breaks assumptions behind traditional “see and hear” verification.

Practical implication: treat audiovisual interaction as a channel, not as identity evidence, for high-risk decisions.

Why approval workflows failed in the Arup case

The payment controls in the incident were designed to validate transaction attributes such as amount, account and approval sequence. They were not designed to answer the upstream question of whether the instruction originated from the real CFO or a fabricated proxy. That distinction matters because financial control frameworks often assume identity has already been established before the transaction review begins. Once the identity input is poisoned, the downstream controls may still function exactly as designed and still approve fraud.

Practical implication: add an independent identity check before the approval workflow begins, not only inside the payment system.

What independent proof of presence changes

Independent proof of presence introduces a second evidentiary layer separate from the conversation itself. In practice, that means the verifier must rely on something other than the live call, such as a controlled out-of-band method, a pre-enrolled trust channel, or cryptographic identity evidence that is hard to imitate. The point is not to make video useless. The point is to stop treating a convincing interaction as sufficient evidence for high-risk authorisation when synthetic media can now replicate the social proof that humans rely on by default.

Practical implication: require a verification path that is logically separate from the conversation being used to request the payment.

Threat narrative

Attacker objective: The objective was to induce a legitimate employee to authorize high-value transfers that appeared to come from senior leadership.

1. Entry occurred through public corporate audio and video that let attackers train convincing deepfake personas without breaching Arup’s core systems.
2. Credential access was social rather than technical: the attackers used a spoofed email and a fabricated video call to obtain authoritative instruction acceptance from the employee.
3. Impact followed when the employee authorized 15 wire transfers totaling HK$200 million, moving funds into five Hong Kong bank accounts before the deception was discovered.

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

Recognition-based approval is now a broken control premise: The Arup case shows that face matching, voice familiarity and a normal video call are no longer reliable proof of identity for high-risk transactions. The issue is not whether the channel works but whether the channel still has evidentiary value when synthetic media can replicate the cues people trust. Practitioners should treat recognition as a weak signal, not an authorization control.

Transaction controls can succeed while identity assurance fails: The company’s financial checks evaluated the transfer, not the provenance of the instruction. That is a classic governance gap in human IAM and payment operations: downstream approval logic can be sound while the upstream identity claim is false. The implication is that transaction governance and identity verification must be designed as separate controls, not one blended process.

Independent proof of presence is becoming a governance requirement, not a nice-to-have: Deepfake fraud creates a named control gap we can call the recognition-to-verification gap. It appears when organizations assume that live interaction is equivalent to verified presence. The practical consequence is that high-risk approvals need evidence that is independent of the conversation itself.

Finance workflow risk is really identity workflow risk: This incident is not just about fraud tactics. It shows that any workflow where human trust unlocks money, data or authority can be compromised once the identity evidence is synthetic. Practitioners should re-map payment, procurement and exception-handling processes as identity assurance journeys, not only as authorization chains.

Cryptographic evidence belongs closer to the approval point: When a request can be convincingly imitated, organizations need stronger proof tied to the person, the request and the session boundary. That aligns with broader digital identity guidance on stronger authentication and with zero trust thinking about continuous verification. The field should stop asking whether people can recognize a face and start asking what verifiable evidence stands behind the request.

What this signals

Deepfake fraud is a human IAM problem, not only a fraud problem: The control failure sits at the point where a person’s judgment is used as identity evidence. That means finance approval, executive escalation and exception handling now need stronger verification steps than traditional awareness training can provide.

Face-based verification has reached its trust ceiling: Once attackers can synthesize credible live audio and video, organizations must assume that presence can be simulated. The governance response is to move verification away from the interaction itself and toward evidence that is harder to imitate.

Recognition-to-verification gap: This incident exposes the gap between believing a familiar person is present and being able to prove that they are. High-risk workflows should be designed so that this gap cannot be closed by appearance alone.

For practitioners

• Separate request and approval channels Require high-risk transactions to be approved in a different system or channel from the one used to request them, so a convincing call cannot carry both persuasion and authorization.
• Add independent verification for executive instructions Use a pre-established callback method, trusted contact list or second approver who was not present on the original call before any material transfer is executed.
• Train finance teams on deepfake escalation scenarios Update fraud and security awareness content so staff can recognise the pattern of urgent email plus familiar video call plus pressure to keep the matter confidential.
• Protect cautious employees from process penalties Make it safe to slow or stop a transaction when the identity evidence is weak, because punitive reactions teach staff to override suspicion instead of verifying it.
• Document proof-of-presence requirements for high-value payments Define which transactions require cryptographic or out-of-band identity evidence, and make the threshold explicit in finance and access governance policy.

Key takeaways

• The Arup deepfake case shows that recognition, not just credentials, can be the weak link in identity-dependent workflows.
• The scale of the loss was US$25.6 million across 15 wire transfers, which shows how costly a successful impersonation can be.
• High-risk approvals need independent proof of presence, because transaction controls alone do not verify who issued the instruction.

Key terms

• Deepfake-Enabled Fraud: Deepfake-enabled fraud uses synthetic media, voice, or face manipulation to impersonate real people and bypass weak verification steps. It raises the bar for identity and fraud teams because traditional checks may not reliably detect fabricated presence, especially in high-volume digital onboarding and support channels.
• Proof of Presence: A verification approach that aims to establish that a real person is actively participating at the moment of authentication. It goes beyond matching a stored trait and instead looks for live, context-specific evidence that resists replay, cloning, and remote fabrication.
• Transactional Authorization: Transactional authorization is permissioning tied to a specific action rather than to broad session access. It lets teams define what an agent may do, where approval is required, and how to prevent a valid session from becoming an open-ended delegation path.
• Identity Assurance: The confidence an organisation has that a person or system is truly who it claims to be before access or action is granted. In modern IAM, assurance depends on evidence quality, channel trust, and the strength of verification around high-risk decisions.

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.