Tackling deepfakes: Let’s focus on authentication, not detection
By Brandon Epstein
This article is part of AI in Digital Forensics, a blog series exploring the impact AI is having in the world of digital investigations.
Key insights
- AI-enabled deepfake detectors are losing the race against deepfake creators.
- Deepfake detectors don’t meet the bar for accuracy that digital forensic authentication tools do.
- Digital investigations should take a page from other fields and focus on establishing authenticity, not just detecting fakes.
- Products like Magnet Verify build trust by analyzing and establishing the provenance of digital evidence.
Cryptocurrency scams portraying well-known people. Fake footage of armed conflict. Fabricated videos of political candidates designed to spread disinformation ahead of elections. These are a few examples of the types of synthetic content that have increased dramatically in the past few years. Synthetic content isn’t all bad. In your social feed, it can be very entertaining, and it’s used in movies and medical training. But synthetic content can be very dangerous when it’s created to mislead people or facilitate criminal activity.
That’s especially true of deepfakes, which are digitally generated or altered images that depict real or fictional individuals or things with the intent to deceive.
The volume of deepfake videos is growing more than 900% annually
Source: Deepstrike
Deepfakes are increasingly easy to create and to scale, making them a concern for governments and organizations vulnerable to fraud, including banks and insurance companies. The legal system, too, is grappling with the problem, given the growing role of video evidence in the courtroom. Because of the widespread dissemination of fakes, there is a new distrust in media evidence.
While public concern tends to focus on scams and fabricated footage circulating online, the more pressing challenge in modern investigations and courtrooms is the reverse: genuine evidence being dismissed as unreliable, and the difficulty of proving authenticity to overcome that doubt.
Current deepfake detectors fall short
It’s natural to assume that the solution lies in establishing what’s real and what isn’t. But several studies indicate that humans are not very good at spotting synthetic content. As it turns out, neither are commercially available algorithmic deepfake detectors.
These detectors may perform well in controlled settings. But there’s growing evidence that in the real world they are unable to reliably and consistently identify deepfakes. And AI content creators are working overtime to outsmart the AI-powered detectors, creating a type of arms race between those trying to make and spot deepfakes.
Rather than spotting the growing number of deepfakes, let’s focus on proving what’s real
In use cases ranging from brand protection to law enforcement, real accuracy is needed. That’s especially true in digital forensics, since we can’t rely on unreliable measures when presenting evidence in court.
We believe a more effective approach to the question of whether digital evidence has been manipulated is to focus on establishing authenticity. Authentication isn’t a new idea.
- Currency issuers have long embedded watermarks in banknotes to reduce counterfeiting.
- Major League Baseball applies tamper-resistant holograms to baseball memorabilia.
- And amid a tsunami of fashion dupes, authentication is also being used to verify the provenance of luxury items.
One effort to formalize the authentication process in the digital media space is being led by the Coalition for Content Provenance and Authenticity, or C2PA. Backed by several major tech companies, news organizations and others, C2PA is promoting digital provenance techniques that would be adopted by parties across the ecosystem to establish content authenticity. While a step in the right direction, C2PA has not proven effective to combat deepfakes, or prove authenticity, in criminal justice or digital investigations.
In the case of digital forensics, authenticating media goes beyond determining whether something is real on a simple yes/no basis. It means using products like Magnet Verify to establish provenance through a comprehensive analysis of a media file’s structure. It involves using verifiable and repeatable methods to identify the source of the file —including the device it originated from — and establish its generational history.
Magnet Verify is a purpose-built media authentication solution for digital investigations. Its automated analysis examines over 100 data points to identify original imagery and detect editing or synthetic creation. Using patented technology that goes beyond metadata and visual inspection, Verify gives investigations unprecedented insight into digital media files — including those transmitted or altered since creation.
This approach not only builds trust in the evidence but is designed from the ground up for courtroom admissibility. What’s more, it makes the job of the human examiner easier, because it allows them to tell a more complete story about the evidence and its history and build confidence in media evidence used in investigations and prosecutions.
Brandon Epstein, Technical Forensics Specialist at Magnet Forensics, is a former police detective and co-founder of Medex Forensics, which Magnet acquired in 2024. Brandon specializes in AI and media authentication and is active in many digital forensic community organizations.
Watch Brandon Epstein’s AI Unpacked webinar series on how responsible AI is shaping the future of digital forensics.