Deepfake Detection Companies: 1,200 Traded Faces Exposed (Updated)

A Telegram room with 1,200 members wasn't just trading fake photos. It was trading addresses, student IDs, and workplace details — the raw ingredients that turn a blurry AI fake into something that could actually fool your mom, your boss, or your ex. That's the part nobody warns you about. This is why deepfake detection companies exist at all: not because AI got scary, but because people got careless with each other's information first.

TL;DR: Deepfake detection companies are getting more attention because Telegram "exposure rooms" show that ordinary people's photos, names, and personal details — shared casually by other people — are what makes AI fakes convincing in the first place.

Here's the thing that should keep you up tonight: it's not your face that's the problem. It's your context. Someone mentions where you work. Someone else drops your relationship status. A third person shares a photo you never posted publicly. None of that is a crime by itself — it's just talk. But feed it into a Telegram room built for this exact purpose, and you've got a targeting kit. The report describes teenagers running "exposure rooms" that operated for seven months, collecting personal information at the request of other members before the fabrication even started. Two-step process: gather first, fake second.

---

Why deepfake detection companies can't fix a problem that starts with gossip

Deepfake detection companies build tools to spot manipulated video, images, and voice after the fact — checking whether a match score (a number showing how alike two faces are) lines up, or whether skin appears too smooth in a way real skin never does. That's genuinely useful. A good deepfake detector can flag a manipulated clip in seconds, and newer detection technology keeps getting better at catching subtle blending errors. But detection is defense, not prevention. It kicks in after your face is already stitched into something fake. The actual vulnerability — the thing nobody's selling a fix for — is the moment before that, when someone who knows you decides your job title, your school, or your relationship status is fair game to share in a group chat full of strangers.

According to reporting compiled by Fortune, South Korean police found a Telegram group with 1,200 members sharing fabricated images alongside home addresses and student ID numbers. Read that again. It wasn't a fake-photo swap meet. It was a targeting database with pictures attached. And according to NBC News, the public outrage that followed pushed South Korea toward a genuine crackdown on Telegram-based deepfake pornography — not a small policy footnote, but a national reckoning over how normal it had become to treat women's personal media as shared property. Protecting media authenticity at scale is exactly what a serious detection company is supposed to help with, even if it can't touch the gossip that starts the chain. Detection vendors also use media samples to train their systems, since every additional clip of manipulated media helps a deepfake detection model learn what tampering actually looks like.

Teenagers arrested for running Telegram exposure rooms had posted personal information of victims at the request of other participants over a seven-month period. This article is part of a series — start with Illinois Bipa Court Says A Recorded Voice Is Now A Face Scan.

— Reporting summarized by Asiae

How does deepfake detection actually work on stolen photos?

Most deepfake detection tools detect tiny mismatches a fake leaves behind — lighting that doesn't quite match, blinking patterns that look off, or blending seams around the jaw. Some check "liveness" (whether a camera is looking at an actual live person, not a photo or a screen). But if the starting photo was already stolen and shared privately before any AI touched it, detection tools can only detect the fake after it exists — they can't undo the fact that your photo was circulating in a room you never knew existed. Every deepfake detector on the market shares this same blind spot, no matter how well it can detect pixel-level tampering.

Where deepfake detection software fits in the bigger picture

Good detection software and browser-level scanning tools are part of a wider push toward content integrity — the idea that a piece of media should carry some proof of where it came from and whether it was altered. A browser extension that flags manipulated video is genuinely helpful for everyday users, but it still can't see the private group chat where your photo was first collected.

---

The Telegram deepfake room problem is bigger than one platform, but Telegram made it easy

Let's not pretend this is unique to one app. But Telegram's structure — big anonymous groups, weak enforcement, bots that do the fabrication work automatically — made it the obvious home base. Investigations reviewed by The Decoder describe a monetized system: bots that strip clothing from photos on command, archives that store and resell the results, and channels that treat this like a business with customers. That's not a rogue teenager with a laptop. That's an economy built on trading deepfake images the same way stolen data gets traded elsewhere.

And it's not slowing down. A January 2026 investigation from Creati.ai found millions of people using AI tools on Telegram to generate non-consensual deepfake nudes — a scale that makes "one bad room" language feel almost quaint. According to NPR, South Korean investigators opened a formal probe into Telegram itself over alleged sexual deepfakes targeting women and minors — a sign regulators finally see the platform, not just the individual poster, as part of the problem.

---

What deepfake detection companies can and can't catch — and where the market is going

There's real momentum behind building better detection. Independent research groups have started ranking tools on how well they catch fabricated video and voice under pressure — and yes, deepfake detection with gartner style evaluation frameworks are starting to show up as companies compete for a spot as the market leader in a crowded field. The pitch from companies in this space is straightforward: run a piece of media through the software, get a security read via an api call on whether it's real, and some vendors now offer a second api tier for bulk media scanning. Some position themselves around enterprise-scale threat response and security operations, while others market tools for identity verification pipelines, and still others — like tools built to identify deepfakes in call centers — focus on voice-based fraud and scam attempts rather than images. Previously in this series: Synthetic Identity Fraud Fake Mahama Video Sold Crypto Scam.

None of that helps the person whose home address just got dropped into a Telegram room, though. That's the gap. Some companies' reliable deepfake detection claims focus on catching manipulated video and images after upload — useful for platforms and governments trying to moderate content at scale, less useful for an individual who has no idea a fake even exists yet. Reality Defender is one name that keeps coming up in these conversations, and reality defender is often cited alongside other detection systems built for enterprise and government security teams rather than individual victims.

What deepfake detection tools do	What Telegram-room targeting does first	Status
Scan uploaded video or photos for signs of AI manipulation using a deepfake detector	Collects your name, workplace, school, or relationship details from people who know you	Ongoing, 2025-2026
Flag mismatched lighting, blending, or blink patterns in a piece of media	Shares real photos of you without your knowledge, often taken secretly	Ongoing, 2025-2026
Give platforms a match score via an api to act on for takedowns	Builds a targeting profile before any fake image is created	Ongoing, 2025-2026
Works after the fabrication already exists, detect flaws in deepfake media after the fact	Works quietly, weeks or months before the fabrication appears	Ongoing, 2025-2026

Does verification stop a stolen photo from becoming a deepfake?

Not on its own. Verification (confirming a photo or account really belongs to the person it claims to) helps platforms catch impersonation accounts and fake profiles. But it doesn't stop someone from screenshotting your real, verified profile photo and feeding it into an AI tool in a private room. Verification protects the front door. It does nothing about the side conversation happening about you two rooms over.

---

Deepfake detection companies won't warn you about a coworker asking for "one more photo"

Here's the thing I actually want you to do something about. If you've ever wondered whether a photo or profile is really who it claims to be, that's the exact question this kind of tech exists to answer — deepfake detection tools and identity verification systems are built for that moment of doubt. But the moment that actually matters happens earlier, and it's not technical at all. It's when someone — a coworker, an ex, a "friend of a friend" — asks for a few extra photos, or casually confirms your workplace, your school, or who you're dating, under the excuse of "just double-checking" or "quick verification." That request is not small talk. Treat it like a flare going up. The people running these Telegram rooms didn't need hackers. They needed chatty acquaintances.

Think about how this plays out with the availability heuristic — the mental shortcut where we judge risk by what's easy to picture, not by what's actually likely. Most people can picture a hacker breaking into a database. Almost nobody pictures their coworker's off-hand comment in a group chat becoming step one of a targeting operation. That gap in imagination is exactly why this keeps working.

---

So here's the question worth sitting with tonight: if someone used your public photos and a few personal details to build an AI fake of you, would the people closest to you actually know how to tell it was false — or would they just believe it, the way 1,200 people in one Telegram room apparently believed whatever showed up in their feed? Up next: Illinois Bipa Court Says A Recorded Voice Is Now A Face Scan.

deepfake detection companies: Frequently Asked Questions