Most fraud companies sell you the same verb: prevention. Ekan Subramanian opened by rejecting it. “It’s not about preventing fraud,” he told me. “We identify.” The distinction sounds like hair-splitting until he explains it, and then it reorganises how you think about the whole problem. His company is called Fingerprint for a reason: just as no two human fingerprints match, Fingerprint assigns a unique, unforgeable identity to every device that visits a website. But identifying the device is only half of it. The half that matters is reading its intent.
| What You Need to Know |
|---|
| 1Interview recorded at Money 20/20 Europe 2026, 2 to 4 June, at the Amsterdam RAI. |
| 2Ekan Subramanian is VP of Engineering at Fingerprint, a device-identification company that began as the open-source FingerprintJS library and now sells device intelligence to fight fraud. |
| 3His reframe: fraud is not about prevention but identification and intent. Identify every device, then decide whether it is a legitimate visitor (make it frictionless) or a fraudster (add friction). |
| 4The engine is scale: 100+ signals and, in Subramanian's framing, 1.5 billion mapped devices let its machine-learning models spot a brand-new fraud pattern instantly. |
| 5Its newest front is AI: just after Money 20/20, Fingerprint launched an Automation Intelligence API to detect AI agents and assistants. |
Subramanian is VP of Engineering at Fingerprint, and we sat down at Money 20/20 Europe as the entire show buzzed about AI agents. That turns out to be the exact problem he is now racing to solve, because in a world where humans, bots and AI agents all show up at your door, you can no longer just ask who is knocking. You have to ask why.
Why you can’t just block the bots
The reason identification alone is not enough is the part most people miss. If all you do is detect non-human traffic and block it, Subramanian said, you go out of business, because some of that traffic is good: helpful bots, legitimate AI agents acting for real customers. So Fingerprint does two things at once. It uniquely identifies each device that arrives, and then it judges intent.
“Are they legitimate customers? Then you don’t want to introduce friction and frustrate them. Are they coming in with a fraudulent intent, trying to commit something? Then we share that information so our customers can add friction to prevent the fraud.”
That second use case, the good customer, is the half that gets forgotten in fraud conversations, and Subramanian was keen to stress it. When Fingerprint recognises a known device as a trusted customer, the goal is to make that person’s journey as seamless as possible. Friction is not free; pile on too many “please verify yourself again” checks and you lose the very customers you wanted, who decide your service is a hassle. Good fraud defence, in his telling, is as much about waving the right people through as it is about stopping the wrong ones.
The engine: 100 signals and a billion-plus devices
How do you tell those two apart at internet scale? Subramanian walked through the machinery. A research team’s whole job is finding the signals, more than 100 of them, that together identify a visitor accurately, and they are tested ruthlessly: internally first (“we eat our own dog food,” running it on Fingerprint’s own site), then slowly with customers, never shipped until the accuracy holds up.
The advantage compounds with scale. Drawing on more than 100 signals across what he framed as 1.5 billion uniquely mapped devices (a figure worth attributing to him; Fingerprint’s published numbers run to over a billion device identifications a month), the company’s machine-learning models have learned what normal looks like. So when a fraudster evolves a new technique, the system can flag that the pattern is strange, something it has not seen across all those devices, and react. Staying ahead also means staying close to the browser makers, Google and Apple, whose changes ripple through everything. The arms race is permanent: “We have to stay ahead of them and be smarter than them,” he said, because that is the only way to keep up with adversaries whose entire job is getting past whatever you built.
The AI front opens
The conversation kept circling back to the year’s obsession, and Subramanian agreed AI is both the biggest opportunity and the biggest threat ahead. We had heard from other guests this week that AI now lets a single fraudster run scams against a hundred people at once, the kind of industrialisation that platforms like Trustfull and Nominis are also racing to counter. Fingerprint’s response is to treat AI agents as a population to be identified rather than a wall to be built.
In February the company released an AI-agent ecosystem that detects agents and begins categorising the good ones, which increasingly announce themselves through an emerging signed protocol. And just days after Money 20/20, Subramanian noted, Fingerprint launched its Automation Intelligence API to detect AI assistants too, so that AI-agent detection becomes part of mapping a customer’s journey before it ever hits the site. The hurdle he sees for the coming years is simply that AI evolves fast, which means his research team’s job is now a permanent watch on new agent patterns mined from the company’s own traffic.
The reason Fingerprint belongs at a money show, he closed, is that fraud refuses to respect industry lines. Game studios, banks, fintechs, acquirers, payment gateways, all of them are in that hall, and all of them get hit, whether by promo abuse, payment fraud, credential stuffing or social engineering. Fraud is universal, which is why a defence built on universal identification travels everywhere money does. He left me with the line his colleagues clearly love, and it earns its place: fraud is fast, but Fingerprint is faster.
This interview was recorded live at Money 20/20 Europe 2026 at the Amsterdam RAI. For more on the fight against fraud, watch the full Cryptofocus series, including our interviews with wallet-risk platform WAML and fraud-prevention firm Trustfull.
