Can AI really extract your fingerprint from a photo?

Not the fingerprint we usually write about

Worth clearing up before anything else, since we cover a completely different kind of "fingerprint" elsewhere on this site: browser fingerprinting identifies your device using signals like fonts and GPU rendering, and has nothing to do with your actual finger — you can see what that looks like with our browser fingerprint test. This article is about the literal, physical biometric — the ridge pattern on your fingertip, the one that can unlock a phone or authorize a payment.

Where this actually started

The underlying finding isn't new, and it isn't a rumor. In 2017, Isao Echizen, a professor at Japan's National Institute of Informatics, demonstrated with fellow researcher Tateo Ogane that fingerprint data could be lifted from an ordinary digital photograph of someone flashing a peace sign, taken from about 3 meters away using a digital SLR with a 135mm lens. Echizen told Reuters at the time that a copied print could plausibly be used to unlock a phone or gain entry somewhere it shouldn't. That result has sat quietly in biometric security research for years, cited occasionally, without becoming a mainstream fear — the equipment involved, a dedicated telephoto lens on a proper camera at a fixed distance, isn't what most photos are taken with.

Why it went viral again in 2026

The current wave traces back to a Chinese television segment where a financial commentator, Li Chang, used a celebrity's peace-sign selfie and image-editing software to sharpen visible ridge detail on camera, presenting it as a live demonstration of the risk. Clips spread well beyond their original audience, and the framing shifted from "a research finding from 2017" to "AI can steal your fingerprint from any selfie," which is a meaningfully different and much scarier claim. Social posts reacting to the clips picked up thousands of likes, several openly panicked about ever posing for a photo again.

What the experts pushed back with

The reaction from biometric security researchers, once the clips reached a wider audience, was consistently more measured than the panic suggested. Justin Cappos, a cybersecurity professor at New York University whose research has been used by companies including Google, put the realistic odds in blunt terms: for most people, he said, the chance of this actually happening is smaller than a better chance of being hit by a car tomorrow. Vyas Sekar, an electrical and computer engineering professor at Carnegie Mellon, made a similar point by comparing the scenario to spy-movie plotting rather than a realistic street-level threat.

A 2025 peer-reviewed paper out of Mbarara University examined the technical feasibility directly, testing whether fingerprint information could actually be recovered from photos as typically shared on social media. Its conclusion split the difference sensibly: the underlying risk is real, but the practical barriers to successfully exploiting it remain significant for the average person's ordinary, casually-taken photos.

The conditions that actually matter

The gap between "technically possible" and "realistic threat to you specifically" comes down to a small number of conditions that all have to line up at once, and casual photos usually fail at least one of them.

Distance and lens

Pei Zhiyong, director of the Qianxin Industry Security Research Centre, has noted that pulling a usable print from a real-world photo is genuinely difficult because it depends on shooting distance being right, not just theoretically possible. Chinese security researcher Li Chang's own stated range puts high recovery likelihood within about 1.5 meters, dropping to roughly half the ridge detail out to 3 meters — a phone selfie at arm's length is well inside the risky range; a group photo across a room generally isn't.

Resolution and compression

This is the condition that quietly protects most people without them doing anything: photos uploaded to Instagram, Facebook, or WhatsApp get compressed, often substantially, which destroys exactly the kind of fine, high-frequency detail a ridge pattern depends on. A photo straight off a phone's camera roll carries far more of that detail intact than the same photo does after it's been through a social platform's upload pipeline.

Focus, lighting, and orientation

The finger pads need to be in sharp focus, evenly lit, and angled toward the lens rather than at a slant. A candid photo with a hand mid-gesture, slightly blurred or backlit, simply doesn't carry the same risk as a deliberately posed, well-lit, in-focus close-up.

Getting from an image to an actual attack

Even a clean, high-resolution capture of a fingerprint isn't automatically a working key to anything. To defeat a physical fingerprint sensor, the pattern generally has to be turned into something with actual texture a sensor can read — researchers studying this have used materials like alginate, the same seaweed-derived substance used in dental molds, to cast a physical, textured fake finger capable of fooling certain sensors. That's a meaningfully higher bar than just having a photo, and it's part of why the 2025 attempted break-in in Hangzhou, where a group tried to unlock a smart door lock using a previously posted photo of the homeowner's hand, was caught and stopped before it succeeded. It demonstrates the theoretical path is real; it doesn't demonstrate that path is easy.

The biometric fraud actually happening at scale

Here's the part that gets lost in the peace-sign panic: the biometric fraud actually being documented at volume right now doesn't look like someone lifting your print from a vacation photo at all. Europol's April 2025 report on emerging fraud identified AI-generated synthetic fingerprints, alongside deepfakes and cloned voices, as active tools being used to defeat biometric authentication systems directly — not by copying a real person's print, but by generating a fake one algorithmically. Separately, Group-IB research published in January 2026 documented 8,065 biometric injection attacks against a single financial institution over an eight-month window — a technique where fraudsters feed a fabricated biometric signal directly into a verification system's software pipeline, bypassing the camera or sensor entirely rather than physically reproducing anyone's print.

What's actually worth doing

None of this means the underlying risk is fake, just that it's narrower and rarer than the viral framing suggested. A few genuinely useful habits, roughly in order of how much they actually matter:

• Avoid deliberately close, sharp, well-lit photos with fingertip pads angled directly at the camera — that's the specific combination the research depends on. An ordinary compressed selfie posted to social media is a different risk profile entirely.
• Think twice before photographing your own hand up close for something like a product listing or a craft photo — that's a higher-resolution, higher-risk context than a casual selfie.
• Treat fingerprint authentication as one layer, not the only one — back it with a second factor for anything that actually matters, since a compromised fingerprint, unlike a password, can't be reset.

Frequently asked questions

Should I stop making a peace sign in photos?

Not necessarily, and treating it as a hard rule misses the point. What matters is the underlying conditions — distance, resolution, focus, and lighting on the fingertip pads — not the specific hand gesture. A peace sign photographed from across a room, then compressed for social media, carries a different risk than a deliberate close-up regardless of which fingers are showing.

Does this mean my old social media photos are already a risk?

For most people, unlikely, specifically because of platform compression. Photos already uploaded and compressed by Instagram, Facebook, or similar platforms have generally lost the fine ridge detail this technique depends on, even if the original, uncompressed file would have carried enough detail.

Is fingerprint login on my phone still safe to use?

Yes, for the realistic threat model most people face. The documented large-scale biometric fraud trend targets software-based identity verification systems directly through synthetic data injection, not physical device sensors, which generally include their own liveness and spoof-detection checks that a flat photograph or printed image doesn't easily pass.

What's the difference between this and the AI-generated fingerprints Europol warned about?

Two different attack paths entirely:

• Photo extraction starts from a real person's actual fingerprint and tries to reproduce it, which requires a real, high-quality photo of that specific person's hand.
• Synthetic fingerprint generation doesn't copy anyone's real print at all; it algorithmically produces a fake one designed to pass a verification system's checks, then feeds it in directly through software rather than a camera.

The second is the one showing up in fraud statistics at meaningful volume right now.

Sources

• Gulf News / Reuters — the original 2017 NII fingerprint extraction research
• CBS News — expert reaction to the 2026 viral panic
• SOFX — Europol's April 2025 report and Group-IB's biometric injection attack data
• The Rakyat Post — the 2025 Hangzhou smart-lock incident
• arXiv — research on physically reproducing fingerprints to defeat sensors