A researcher at Palo Alto Networks’ Unit 42 sat down at a 5-year-old computer with no image manipulation experience and no deepfake knowledge. Seventy minutes later, that researcher had built a synthetic identity convincing enough to pass a live video job interview. That’s the quick on-ramp for creating a deepfake that attackers capitalize on: going from no experience to a passable deepfake in just over an hour.
That experiment was designed to replicate what North Korean state-backed operatives are doing at industrial scale, where stolen credentials and fabricated personas are combined into complete synthetic identities.
The lab proved something worse about the tools themselves: They are so accessible that any motivated attacker can do it from a bedroom. Enterprise identity verification was supposed to stop exactly this, but it didn't. And the enterprises getting hit keep filing the losses under “fraud” instead of recognizing what they are: identity-layer failures that no anti-fraud tool was built to catch.
The numbers no longer describe a future threat
Generative AI-enabled fraud losses in the U.S. will climb from $12.3 billion in 2023 to $40 billion by 2027, attaining a compound annual growth rate of 32%, according to projections from the Deloitte Center for Financial Services. That same report documented an entire illicit industry on the dark web selling scamming software for as little as $20 to thousands of dollars.
On November 13, 2024, FinCEN issued a formal alert warning financial institutions of rising deepfake fraud in suspicious activity reports. The agency had observed a sustained increase in SARs describing the use of deepfake media to circumvent identity verification and authentication methods.
The real vulnerability is organizational, not technical
Here is the uncomfortable truth most security leaders have not yet confronted: Deepfake-driven identity abuse has no home in their budgets. Incidents get categorized as “fraud” and routed to anti-fraud teams whose controls were designed for stolen credit cards, not AI-generated video of the CFO authorizing a wire transfer.
Identity verification sits with compliance. Voice authentication belongs to the contact center. Videoconferencing security lives with IT operations. Nobody owns the identity layer as a unified attack surface, and the gap sits between budget lines.
Mike Riemer, Ivanti’s Field CISO, has watched this pattern repeat across two decades of security frameworks. “I’ve been through defense in depth, least privilege, and now zero trust. It all comes down to: do you know who is on the other side of the keyboard?” Riemer told VentureBeat. “Until I can validate that, I’m not going to communicate with them. But most enterprises still treat identity verification as something that happens once at the front gate and never again.”
Gartner predicted that by this year, 30% of enterprises will no longer consider standalone identity verification and authentication solutions reliable in isolation due to AI-generated deepfakes. Injection attacks on biometric systems increased 200% in 2023, and current presentation attack detection (PAD) standards do not cover them.
Five identity-layer blind spots attackers are exploiting now
1. Biometric verification assumes the camera is honest
Most enterprise identity verification relies on presentation attack detection, which checks whether a real human face is in front of a camera. Digital injection attacks bypass the camera entirely, feeding synthetic video directly into the verification system’s data stream. PAD standards do not cover AI-generated deepfakes that exist now, and liveness detection never fires.
2. Voice authentication trusts the wrong signal
Voice cloning now requires as little as a few seconds of source audio, according to multiple commercial platforms, including OpenAI’s Voice Engine, which demonstrated cloning from a 15-second clip. Pindrop’s analysis revealed a 756% year-over-year increase in deepfake or replayed voices across enterprise phone calls. Yet voice-based authentication remains a trusted factor in financial services helpdesks and executive approval workflows. The signal that validates a voice sound familiar is the exact signal that attackers can now fabricate.
Separately, Sumsub’s 2023 Identity Fraud Report found deepfake incidents in fintech grew 700% year-over-year.
“If they want to compromise someone’s identity, they call the help desk and pretend to be that person,” Adam Meyers, Senior Vice President of Counter Adversary Operations at CrowdStrike, told VentureBeat. “They’ve done the open source information collection; they know everything they need to know. They call the help desk and say, ‘Hey, I can’t log in.’ And that’s how it starts.”
The attack surface is not an authentication system. It is a human on the other end of the help desk line, operating on trust and a familiar voice.
3. Video conferencing is treated as an authenticated channel
The Arup attack succeeded because video calls carry implicit trust. A familiar face on a scheduled platform triggers every verification instinct to say, “This is real.” Enterprises built their remote-work approval workflows on that assumption. That assumption collapsed, and most security teams have not caught up. GetReal Security's Deepfake Readiness Benchmark Report found that 88% of organizations now encounter deepfake or impersonation attacks at least occasionally, and 41% have hired and onboarded a fraudulent candidate.
4. Remote hiring pipelines rely on visual trust that AI has rendered worthless
North Korean state-backed operatives have infiltrated more than 320 companies in the past 12 months using stolen identities and real-time deepfake technology during video interviews, according to CrowdStrike’s 2026 Global Threat Report. The number of companies that unknowingly hired North Korean developers grew 220% year-over-year. The Unit 42 experiment described above proved that the barrier to entry is gone. A bedroom, a commodity laptop, and an hour of effort produce a synthetic persona that defeats most hiring pipelines.
Even KnowBe4, a security awareness firm, unknowingly hired a North Korean operative in 2024.
5. Human detection fails 75% of the time against high-quality deepfakes
Peer-reviewed research shows humans identify high-quality deepfakes as fakes only 24.5% of the time, even when warned. Ironscales’ Fall 2025 Threat Report found that over half of surveyed organizations reported financial losses tied to deepfake or AI voice fraud in the past year, averaging over $280,000 per incident. Nearly 20% reported losses of $500,000 or more. That gap between perceived readiness and actual losses is where attackers live.
Nobody owns the identity layer
These five blind spots share a common root cause. Identity verification, voice authentication, video-based trust, hiring validation, and human detection all belong to different teams with different budgets reporting to different executives. The CISO owns endpoint security. The chief compliance officer owns KYC. The VP of HR owns hiring. The CFO owns financial controls. Deepfake-driven identity abuse does not respect those boundaries. It exploits the seams between them. That is the vulnerability no vendor patch can fix.
Kayne McGladrey, IEEE Senior Member, told VentureBeat that the ownership gap is structural. “The CISO doesn’t have the budget. The CISO doesn’t have the staff. And even if they had the budget and the staff, they wouldn’t have the project management and the change management departments reporting to them” to execute identity controls across every business function. The CISO can observe and advise on identity risk, but the business systems where identity fraud occurs belong to other leaders.
The industry recognized this exact problem four years ago. Gartner introduced Identity Threat Detection and Response (ITDR) in 2022 as a dedicated security discipline, defining it as the tools and best practices that protect identity infrastructure itself from attacks. Gartner Research Vice President Peter Firstbrook said at the time that organizations had poured effort into IAM capabilities, but most of it focused on user authentication, which actually increased the attack surface. ITDR was built to close the gap McGladrey describes: a unified discipline that bridges the IAM and SOC silos into what Gartner now calls a fusion team.
Enterprises got the message. Siemens deployed ITDR as part of its cloud-first zero trust posture. Thomas Mueller-Lynch, Siemens’ Service Owner Lead for Digital Identity, stated publicly that the company now has real-time visibility to defend against attacks. It is a production deployment by an enterprise that decided the identity layer was worth defending as infrastructure.
The Gartner Peer Insights ITDR market now lists rated vendors with enterprise reviews. Four years after the category was defined, the discipline exists, the tools exist, and the enterprises running them have receipts that tally their results repelling attacks. Most organizations still have not deployed ITDR with the cross-functional authority it requires. The ownership gap McGladrey describes is the reason.
Reclassifying the threat changes the defense
The enterprises that will survive this shift are the ones that stop calling it fraud and start calling it what it is: identity-layer compromise. That reclassification moves the budget from fragmented anti-fraud tools into a unified identity infrastructure. It forces the CISO, CCO, and CHRO into the same planning cycle. It makes continuous identity verification the architectural standard.
Detection alone cannot close a gap that is fundamentally organizational: one owner, one budget line, every channel treated as a deepfake delivery mechanism.
What to do Monday morning
The five blind spots above share a single architectural failure: No one person owns the identity layer, so no one defends it. These five maps directly address those gaps. None requires new technology. All require someone to decide that the identity layer is their problem.
Name the owner. One executive accountable for synthetic identity risk across security, compliance, HR, and finance. Not a committee. If your org chart cannot answer “who owns deepfake defense,” that is your first vulnerability.
Kill every single-channel approval. Any workflow where video alone, voice alone, or email alone can authorize a financial transaction is compromised. Require a second verification channel for every action above your risk threshold, using a callback to your corporate directory, not the number the caller provided.
Require injection attack detection from every IDV vendor. Presentation attack detection is insufficient. Gartner’s research is unambiguous: injection attacks bypass the camera entirely. Before renewing any identity verification contract, require vendors to demonstrate injection attack detection and image inspection in a live environment. Demand continuous identity signals, not point-in-time checks. If your current vendor cannot clear that bar, that is your answer.
Reclassify deepfake defense in the budget. Move it from anti-fraud to identity security. Create a dedicated line item covering continuous identity verification, deepfake detection, and cross-channel signal correlation. This threat gets underfunded when categorized as SOC tooling. It becomes fundable when framed as a digital identity risk.
Mandatory holds on high-value transactions. No exceptions. Twenty-four-hour holds on wire transfers above your defined threshold. No exceptions for urgency or seniority. Every deepfake attack depends on time pressure. A hold collapses that leverage. Arup’s 15 transfers moved in a single day.
Thirty days for the owner, single channel kill, and budget reclassification. Sixty for the vendor audit and transaction hold policy. The Unit 42 researcher built a synthetic identity in 70 minutes. The toolkit costs less than dinner. Attackers are not waiting.