Face value: What it takes to fool facial recognition

Facial recognition is increasingly embedded in everything from airport boarding gates to bank onboarding flows. The widely-held assumption is that a face is hard to fake and that matching a live face to a trusted source is a reliable identity signal.
Jake Moore, ESET Global Cybersecurity Advisor, recently put this assumption through several practical stress tests. His experiments showed that the powerful technology can actually be both misused and defeated.
In one test, Jake used a pair of modified off-the-shelf smart glasses that can identify people in real time. He walked through a public space, captured people’s faces and compared them against publicly available online data sources, with identity matches returned within seconds. The names and social media profiles were pulled from nothing more than people’s glances.
This ability might come in handy if, say, a conference attendee struggles to remember people's names, but it’s far less palatable when you consider what someone with ill intentions could do with that information.
The second demo had a different spin. It went after financial services, turning a fraud prevention system against itself. Using AI-generated images and freely available software, Jake created a fictitious face to open an actual bank account. The bank's facial recognition and eKYC (know your customer) platform accepted it as a genuine person.
After proving the point, Jake closed the account and shared all information with the bank, which has since shut down that specific method of identity abuse. But one broader question remains: how many financial institutions may still be susceptible to this kind of attack?
Lastly, Jake added himself to a facial recognition watchlist at a busy train station in London. He then walked through the monitored area while running real-time face swap software that overlaid Tom Cruise’s likeness onto Jake’s own in the camera feed. The system, which is also used by the UK police, never recognized or flagged him. It was as if he simply wasn't there and anyone actively searching for him on CCTV would have seen the actor instead.
There's a lot more to these experiments than we can cover here – they’re all part of Jake’s talk at RSAC 2026, which is due in San Francisco from March 23rd-26th, 2026. If you're at the conference, consider attending the talk – after all, seeing this all work against an in-production system in a live environment is different from ‘just’ reading about it. To learn more, including about other ESET talks at the conference, visit this website.
The big picture
Facial recognition systems are being deployed with implicit trust that doesn't match their actual resilience when someone tries to break them – even where they only use off-the-shelf consumer hardware and easily available software, just like Jake did. Identity verification that is solely dependent on a face match clearly carries more risk than most people and organizations realize.
The experiments also send a message to vendors of facial recognition systems and anyone responsible for identity verification systems. Among other things, the systems should be tested in attack simulation settings and under other adversarial conditions. The technology behind facial recognition is fragile in ways that matter when someone attempts to subvert it.

Facts Only

* Jake Moore conducted experiments on facial recognition technology.
* Modified smart glasses were used to identify people in real time.
* Matches were returned within seconds using publicly available online data.
* A fictitious face was created to open a bank account.
* The bank's facial recognition system accepted the fictitious face.
* The bank closed the account after the test.
* Jake added himself to a train station’s facial recognition watchlist.
* Tom Cruise’s likeness was overlaid onto Jake’s face during the experiment.
* The system failed to recognize Jake.
* The experiment took place at a busy train station in London.
* The UK police also use a similar system.
* The experiment was part of a talk at RSAC 2026 (due March 23rd-26th, 2026).

Executive Summary

Facial recognition technology is increasingly vulnerable to manipulation, as demonstrated by experiments conducted by ESET Cybersecurity Advisor, Jake Moore. The tests revealed that readily available consumer hardware and software can be used to spoof facial recognition systems, raising significant concerns about the reliability of this technology as a means of identity verification. Specifically, Moore successfully mimicked a person's face in real-time using modified smart glasses and AI-generated images, demonstrating the ability to access accounts and bypass security protocols. These findings challenge the prevalent assumption that facial recognition is inherently secure and highlight the potential for misuse. The bank’s response, while proactive in shutting down a specific fraud attempt, underscores the broader systemic vulnerability of financial institutions and other organizations relying on facial recognition for authentication. The experiments also demonstrate the fragility of facial recognition systems when faced with adversarial conditions, suggesting that further testing and security enhancements are urgently needed.

Full Take

The article presents a significant challenge to the growing reliance on facial recognition for security and authentication. Moore’s experiments reveal a fundamental vulnerability: the technology is susceptible to manipulation by relatively simple means – consumer-grade hardware and readily available software. This isn't a sophisticated hack, but a demonstration of how easily the system's core assumption – that a face is unique and unforgeable – can be undermined. The “smart glasses” test highlights the risk of identity theft through readily accessible data, while the bank account fraud demonstrates a targeted vulnerability within a seemingly secure system. This narrative plays into the ARC-0043 Motte-and-Bailey pattern – the article initially presents a seemingly straightforward demonstration of a vulnerability, then subtly reinforces the inherent weakness of the system itself ("even where they only use off-the-shelf consumer hardware and easily available software"). The experiment at the train station further amplifies this, creating an almost surreal scenario where a recognizable celebrity image effectively bypasses a legitimate security system. The overall paradigm driving this narrative is a distrust of technological ‘solutions’ – particularly those marketed as ‘secure’ – without rigorous, adversarial testing. The implications are profound: the assumption of inherent trust in facial recognition systems may be dangerously flawed, potentially jeopardizing sensitive data and undermining trust in established institutions. This resonates with ARC-0024 Ambiguity - the precise boundaries of this vulnerability are unclear, highlighting the difficulty in predicting and mitigating risks associated with rapidly evolving AI technologies. The bank's reactive response, while a positive step, suggests a systemic problem – a willingness to accept ‘proof of concept’ demonstrations as sufficient justification for system overhaul. What’s missing is a broader discussion about the ethical implications of deploying a technology that inherently relies on surveillance and raises serious concerns about privacy and potential for abuse. The question isn’t just about fixing the system; it’s about questioning the fundamental premise of using a face as a reliable form of identification.

Sentinel — Likely Human

Confidence

This article presents a series of experiments demonstrating vulnerabilities in facial recognition technology, highlighting the risks associated with relying solely on facial matches for identity verification. While the content is well-structured and informative, the reliance on rhetorical devices and promotional elements leans towards human-generated reporting rather than a purely synthetic creation.

Signals Detected

Sentence length variance is moderate, exhibiting a mix of short and longer sentences, consistent with human writing patterns.

The text presents a balanced framing of the issue, typical of journalistic reporting, rather than a deeply persuasive argument.

Reliance on phrases like 'one could argue,' 'it's important to remember,' and 'experts say' indicates a reliance on established rhetorical patterns.

The description of using ‘off-the-shelf consumer hardware and easily available software’ is common in reporting about vulnerabilities, but lacks specific technical detail.

Human Indicators

The inclusion of a specific conference name (RSAC 2026) and dates suggests a promotional element rather than purely analytical reporting.

The repeated emphasis on 'risk' and 'vulnerable' aligns with a common approach to conveying security concerns.