According to DeepStrike, deepfakes multiplied 16-fold in two years, from about 500,000 in 2023 to 8 million in 2025.
Fake AI images and videos are being used to spread political disinformation, impersonate people, and even defraud online platforms and businesses.
Not only are deepfakes becoming more realistic, but they’re also getting easier to create and share online. But companies are fighting back against synthetic content.
Where AI Detection Began
AI content detection first emerged in 2023, shortly after the release of ChatGPT. ZeroGPT, GPTzero, and Originality were among the first to release AI-detection software, but they were designed solely to detect AI-generated text.
Although text detection worked in some cases, the limitation of word-based content led to occasional inaccuracies and drew skepticism. Shortly after text detection emerged, a new type of AI-writing model, called a “humanizer,” emerged too. The humanizer was created to mimic human writing styles and to score higher on detectors as human-written.
Henry Cooper, an IT and security researcher based in New Zealand, says text detection is nuanced.
“[Text detection] isn’t perfect, but it’s also not completely useless.” In cases of standard texts generated by a typical large language model, Cooper says mainstream detectors can typically identify them accurately faster than a human. “There are still margins for error, though. It’s probability-based,” he says.
Although AI-generated text can be harmful, the harm it causes pales in comparison to that posed by deepfakes. Now, when it comes to detecting AI-generated content, the focus is on deepfake videos and images.
Deepfake Detection Goes Deeper
Unlike text detection, analyzing AI-generated images and video content involves examining more data. A deepfake image has millions of pixels to analyze. Those pixels have to remain consistent. A video (being a sequence of images) can have up to 8 million pixels.
Beyond analyzing the pixels, there is also metadata and watermarks to look at. When it comes to videos, these factors exist, but they also tend to introduce more visual or auditory artifacts.
Deepfake detection relies on looking for inconsistencies that are hard for generative models to maintain across an entire image or video. Inconsistencies like unnatural skin texture, mismatched lighting, distorted reflections, or garbled signage, irregular blinking, blurred facial edges, or small differences between facial movement and speech.
For video, detectors can also analyze frame-to-frame consistency, because a face might look realistic in one frame but then oddly shift in shape, texture, or expression in the next.
Another layer would be audio. Synthetic voices can leave behind clues in tone, breathing patterns, pronunciation, or syncing up mouth movements. This makes deepfake detection a multimodal process, combining all visual, audio, and technical signals rather than relying on a single clue. That’s what gives it more reliability than text detection.
The Companies Fighting Deepfakes
The deepfake problem is here. A need to solve it exists. There’s a clear market for deepfake detection, especially for organizations and online platforms. As the problem has worsened, these companies are making progress in trying to solve it.
Reality Defender: Funded
Reality Defender is one of several companies developing tools to detect AI-generated and manipulated media. The New York-based company, founded in 2021, offers deepfake-detection software for organizations such as enterprises, governments, and online platforms, with products that analyze audio, video, images, and text for signs of synthetic or altered content.
Its system is offered through tools such as an API and a web application, which are designed to provide risk scoring and review signals rather than relying on a single indicator. In 2023, TechCrunch reported that Reality Defender raised $15 million in Series A funding led by DCVC, with participation from investors including Comcast and Nat Friedman’s AI Grant.
In 2026, Biometric Update reported that Reality Defender had reached an agreement to provide deepfake-detection capabilities to Orange Business, the enterprise division of the French telecommunications company Orange.
TruthScan: Bootstrapped
Another serious market contender, TruthScan, is an AI detection platform that analyzes images, audio, and video for signs of AI generation. It was developed by the team behind Undetectable AI, which describes TruthScan as “a sister organization.”
When contacted, a spokesperson for TruthScan said that data gained from Undetectable AI “Has been key in providing elements of detection other software cannot and helping fight AI fraud.” The Undetectable AI website states that subscription revenue helps fund TruthScan’s deepfake-detection work.
TruthScan has been cited by fact-checking and media organizations like Snopes in investigations of AI-generated media.
In March, Euronews’ fact-checking team, The Cube, reported that data from TruthScan identified more than 200 AI-generated videos depicting U.S. Immigration and Customs Enforcement agents, many uploaded by a single Instagram account and with similar 10- or 15-second durations.
In another media test, The Signal selected TruthScan as one of five image detectors used to analyze promotional material from Halifax businesses after first testing 20 AI image detectors on a small sample of real and AI-generated images. Fact-checking outlet Snopes recently used TruthScan to analyze a viral image.
Last year, TruthScan’s AI image detection software was integrated into AI detection platforms like ZeroGPT and DeepAI. In February, TruthScan acquired Imagedetector.
DeepStrike: Research, Not Detection
DeepStrike isn’t a deepfake-detection company. It’s a cybersecurity firm that specializes in penetration testing. Founded in 2016 and based in San Francisco, the company employs ethical hackers who simulate cyber attacks against client systems. Its founders are from the bug bounty world.
DeepStrike’s connection to the deepfake space is through its research. The firm publishes data on deepfake volume and fraud trends through its blog. Its most cited finding — that online deepfakes grew from about 500,000 in 2023 to 8 million in 2025 — has been referenced by academics, journalists, and competing security firms.
A computer science researcher at the University at Buffalo cited the figure in a February 2026 article published by The Conversation. Modulate, a voice-detection company, cited it on its product page.
DeepStrike does not sell deepfake-detection software. Its services focus on offensive security: manual penetration tests, red-team exercises, and social-engineering simulations, including phone-based attacks that test whether employees can identify AI-cloned voices. The company offers these through a cloud-based platform it calls Pentesting-as-a-Service.
Its role in the deepfake space is narrow but specific. DeepStrike measures the problem more so, and companies like TruthScan and Reality Defender build tools to solve it.
Given the increase in deepfake-related risks, the market for solutions to prevent them is growing. From AI-attack training and literacy to content analysis tools, companies are fighting back in a bid to mitigate the threats posed by deepfakes.
