New spying threats force rethink of biometric identity checks

The Five Eyes intelligence alliance’s warning this month that Chinese intelligence services are using fake recruiters on LinkedIn and other job platforms to cultivate people with access to sensitive information is a reminder that the most consequential security failures often begin before a system ever performs a check.

Made up of Australian, Canadian, New Zealand, UK, and U.S. intelligence agencies, their joint bulletin, Safeguarding Our Secrets, describes how Chinese intelligence officers or their affiliates pose as recruiters, consultants, and representatives of credible-appearing companies to identify people with access to government, military, economic, or policy information.

Publicly available information can also be coupled to the de-anonymization of information obtained through data brokers. De-anonymization occurs when data that has been stripped of direct identifiers, such as names or email addresses, is combined with other datasets to re-identify individuals.

All of which is designed to turn a real person into an insider. A job offer provides the opening. A remote interview creates rapport, and the request for something like a trial analytical report tests the target’s willingness to provide information. The demands then become more sensitive.

But the episode also illustrates a wider problem emerging across digital identity systems.

Security may be built around sophisticated facial recognition, document authentication, and biometric matching, yet these tools can do only so much when the person, the credential, or the digital stream reaching the system has already been manipulated.

A biometric scanner can compare a face with a stored image, determine whether a fingerprint resembles a template on file, and can confirm that the person in front of a camera resembles the person associated with a passport or account, but what it cannot automatically establish is whether the identity entered into the system was genuine at the outset.

That is the paradox confronting governments, banks, employers, and technology companies that have spent billions deploying biometric identification systems.

The scanner may work exactly as intended while the identity it is asked to validate has been fabricated, blended from genuine information, or taken over by an impostor before the verification process begins.

The Five Eyes warning shows how an attacker does not have to breach a classified database or defeat a secure facility. Instead, the attacker builds a false relationship that causes a real person to cross the security perimeter voluntarily.

The digital equivalent is increasingly visible in remote hiring, online financial services, account recovery, and government identity programs.

A fraudster may create a plausible work history, establish a false email address, obtain or manipulate identity documents, and build a digital persona that appears credible across the systems designed to evaluate it. The deception can then reach the biometric layer.

A LinkedIn profile or other open source information on targets of foreign adversaries can play a role in that process. A photograph, video, work history, employer name, and professional network can give an attacker the raw material to construct a convincing impersonation or a broader synthetic persona.

While a photograph or video snippet alone does not create a complete identity and should not be treated as sufficient evidence of one, it nevertheless can supply the visual source material for an AI-generated image or video when combined with other publicly available or stolen data.

The next step can be a biometric injection attack, as highlighted in Biometric Update and Goode Intelligence’s 2026 Injection Attack Detection Market Report. Rather than placing a photograph in front of a camera, an attacker can manipulate the software path between the device and the identity verification service.

A virtual camera, emulator, or tampered application can submit synthetic video directly into the verification process. The system may receive a moving face that appears responsive and lifelike even though no genuine camera is observing a real person in real time.

This is different from the older presentation attacks that biometric systems were designed to detect.

A printed photograph, a replayed video, or a mask attempts to fool a sensor from the outside while an injection attack can bypass the sensor altogether. It inserts manipulated media into the stream before the system can judge whether the image came from a live camera.

The result is a more fundamental challenge than a bad facial match. An identity verification system can accurately compare a synthetic face with a synthetic record and can validate a selfie against a manipulated document. It can even detect certain signs of movement while failing to recognize that the video itself was generated, altered, or supplied through a virtual device.

This is why a warning from Thales about AI becoming a new insider threat is directly relevant to the Five Eyes bulletin. Thales’ concern, outlined in its 2026 Thales Data Threat Report, is that AI systems and agents are being given access to sensitive data in ways that can make them function like insiders without adequate controls.

The connection is the trust relationship. In one case, a foreign intelligence service uses a false recruiter to cultivate a real employee. In another, an organization gives an AI system access to data and systems that may exceed the controls applied to an ordinary user. In a third, a synthetic applicant gains access to a company by passing through a remote hiring process built on weak assumptions about the person behind the camera.

Each case begins with the same question. Who, or what, is being trusted?

That question is becoming more urgent as work itself moves online. Gartner predicted that by 2028, about one in four candidate profiles worldwide could be fake.

The projection is not about ordinary résumé exaggeration. It reflects a market in which an applicant may be represented by a synthetic identity, a deepfake during a video interview, or another person standing in for the person on the application.

The threat is not limited to a bad hiring decision. A fraudulent employee who gains access to a corporate network, source code repository, customer database, payment system, or privileged administrative tool becomes an insider problem. The identity failure can then turn into a cybersecurity breach, an espionage risk, or a fraud operation.

This has created a new two-sided security problem for employers. The candidate must be verified, but so must the recruiter and the company claiming to offer the work. A system that asks a job applicant for sensitive information, a writing sample, or access to a supposedly secure communication channel may itself be the attack surface.

The biometric market is responding by moving beyond the narrow question of whether a face matches a document. Increasingly, the goal is to establish whether a person is physically present, whether the image stream came from a trusted device, and whether the session shows signs of manipulation.

Liveness detection remains central to that effort, but it is no longer enough to ask whether a face appears to blink, move, or respond to a challenge.

Systems are being designed to examine the integrity of the digital stream itself. They look for evidence that video came through a virtual camera, an emulator, or a compromised application, and they examine metadata, device signals, and irregularities in the feed that can reveal whether the apparent live session was constructed elsewhere.

This has given rise to a separate market for injection attack detection. The technology is intended to determine whether biometric data reaching a system originated with a real sensor in a real capture environment.

It is a recognition that a biometric match has little value if the biometric sample itself has been compromised before the matching process begins.

Deepfake detection has become another layer, as detailed in Biometric Update and Goode Intelligence’s 2025 Deepfake Detection Market Report.

These tools analyze faces, voices, and video patterns for signs that media was generated or altered. Some are designed for enrollment and identity proofing. Others are aimed at remote interviews, customer-service calls, financial transactions, and account recovery.

The most significant shift, however, may come after the person has already been authenticated.

Identity companies are increasingly marketing continuous assurance tools that monitor whether the verified user remains the same trusted user throughout a session and across repeated access attempts.

These systems may evaluate how a device is handled, how a person types, how they navigate an application, or whether their network and device posture suddenly changes. A deviation from the expected pattern can trigger a new verification request.

That approach reflects an uncomfortable conclusion. Authentication cannot always be treated as a one-time event. A face match at login does not guarantee that the same authorized person remains at the keyboard, that the account has not been taken over, or that an apparently valid employee is not operating under coercion or deception.

It also raises difficult privacy questions. Continuous identity assurance can reduce fraud by detecting an impostor after login, but it requires organizations to collect and analyze more information about how people behave while using digital systems.

The effort to make identity more trustworthy may therefore expand the surveillance capabilities built into ordinary employment and consumer platforms.

The Five Eyes warning is ultimately about espionage, but it points to a broader erosion of the line between social engineering and identity fraud. A fake recruiter can recruit a real insider. A fake candidate can enter an organization. A deepfake can impersonate either one, and an AI agent can be granted access as though it were a trusted employee.

The security perimeter is no longer only a gate, a passport reader, or a facial-recognition camera. It begins much earlier, with the creation of the identity, the credibility of the relationship, and the integrity of the digital stream.

Unless those elements can be verified, the most sophisticated scanner may simply authenticate a carefully manufactured lie.

Article Topics

biometrics  |  China  |  deepfake detection  |  Five Eyes  |  identity intelligence  |  identity verification  |  injection attack detection  |  synthetic identity fraud