A Different Kind of Deepfake Defense

So here's the thing about most deepfake detection technology out there right now — it's essentially reactive. Someone creates a fake image of you, it spreads online, and then, if you're lucky, a detection tool eventually flags it. By that point, the damage is often already done. A Korean AI security startup called Stillcut wants to flip that entire model on its head.

Instead of chasing down synthetic images after the fact, Stillcut's approach is to protect the original photo before it can ever be used to generate a deepfake in the first place. And that idea just earned the company a seed investment from Seoul-based venture capital firm Mashup Ventures — along with a growing list of real-world proof points that suggest this isn't just theory.

How the Technology Actually Works

What's really interesting is the mechanics behind Stillcut's core product, called Stillcut Protect. The system embeds an imperceptible signal directly into a photo — something completely invisible to the human eye that doesn't degrade the image quality in any noticeable way. But when a deepfake generation model tries to use that photo as source material, the embedded signal disrupts the synthesis process and blocks the fake from being created cleanly.

Think of it like a watermark, but one that doesn't just identify a photo — it actively fights back.

The company pairs this with a second product, Stillcut Detect, which handles the more conventional side of things: identifying AI-generated or manipulated images and videos after the fact. Together, the two tools cover both ends of the threat — prevention before misuse and detection after distribution.

From Research to the Real World

One of the most compelling parts of Stillcut's pitch is that this technology has already been tested against actual deepfake sites — not just in a lab. The team verified their defensive performance against more than 30 currently operating deepfake platforms. That's not a controlled experiment; that's going up against the real thing.

Perhaps the most visible real-world deployment so far? A partnership with Seoul National University (SNU), one of South Korea's most prestigious institutions. Stillcut has already applied its protection technology to all 5,800 graduation photos from SNU's class of 2026. Given that South Korea experienced a deeply disturbing wave of deepfake sexual exploitation crimes in 2024 — which disproportionately targeted young women using photos lifted from social media — protecting a graduation photo archive like this carries real social weight.

The company is now in discussions to expand its reach to domestic broadcasters and major law firms, sectors where the authenticity and integrity of visual content can have serious legal and reputational consequences. Longer term, Stillcut is exploring applications across broadcast content, corporate documentation, and legal evidence imagery.

The Team Behind It

Stillcut was founded by three Seoul National University alumni, and the founding story has a personal dimension that gives it some grounding. CEO Lee Jung-hoon studied Computer Science and Economics at SNU's School of Liberal Studies and later worked on model development and service operations at an AI semiconductor company. He also had a hobby: portrait photography.

When South Korea's deepfake crisis made headlines in 2024, Lee found himself thinking about a gap that wasn't being addressed — the fact that ordinary people had almost no way to protect their own photos from being weaponized. That observation became the foundation for Stillcut.

He brought in Yoo Hyun-jin as Chief Technology Officer and Ha Kyu-won as Chief Operating Officer, and the three co-founders have been building the company ever since.

Early Validation and What's Next

Beyond the Mashup Ventures seed investment, Stillcut has been racking up external validation at a notable pace. The company has filed five patents across South Korea, the United States, and Europe. In May, it won the finals of the Select USA Tech World Pitching Competition, an event organized by the U.S. Department of Commerce — which is a meaningful signal for a young Korean startup with global ambitions. Stillcut was also selected for the Korean government's TIPS program, administered by the Ministry of SMEs and Startups, through a recommendation from Mashup Ventures. TIPS, which stands for Tech Incubator Program for Startups, provides government-backed R&D funding to high-potential early-stage companies.

Through that program, Stillcut will have access to up to 800 million Korean won — roughly $580,000 USD — to further develop its defense and detection capabilities and push toward acquiring clients both in Korea and internationally.

Why Investors Are Paying Attention

"As AI regulations are implemented in major countries, the demand to verify the authenticity of content is increasing. We highly valued the potential for deepfake prevention processing to become a fundamental feature of digital images in the long term, as well as the capabilities of the team to implement it." — Park Eun-woo, Vice President, Mashup Ventures

That quote gets at something broader that's worth sitting with. Right now, deepfake prevention is still a niche concern addressed by niche tools. But if AI-generated content regulation continues tightening globally — and most signs suggest it will — the ability to verify whether an image is authentic could become as standard as a security certificate on a website. That's the long-term bet Mashup Ventures is making here.

"Most deepfake countermeasure technologies focus on detecting pre-existing results. Now that we have verified the performance of our preemptive defense and detection technologies in commercial environments, we will focus on applying them to domestic and international clients." — Lee Jung-hoon, CEO, Stillcut

So here's the bigger picture: Stillcut isn't just building a product — it's making an argument about where the entire category of AI image security needs to go. Catching fakes after they spread is a losing game. Getting ahead of the generation process entirely is the more defensible position. Whether the market agrees, and whether Stillcut can scale its technology to meet that demand, is the question that will define the next chapter for this team.

This article is based on reports from Venturesquare, Naver News, Yonhap News.