Deepfakes are reshaping truth faster than defenses can

deepfakes are – Digital forensics expert Hany Farid says today’s deepfakes can be created without skill, making it increasingly hard for everyday people to tell what’s real. Farid, a former University of California, Berkeley professor and co-founder of the startup GetReal, wa

For Hany Farid, the hardest part isn’t that fake media exists. It’s how quickly people—especially those just scrolling—are being pulled into a world where truth is no longer easy to recognize.

Farid. a digital forensics expert who previously taught at the University of California. Berkeley and co-founded the digital forensics and cybersecurity startup GetReal. says the average person on the internet today cannot reliably tell whether an image. video. or audio recording is real or not. He points to studies of perception—showing that human visual and auditory systems are simply not good enough to perform that task on their own.

That gap between what technology can produce and what people can verify is widening, he argues. “You give us a piece of content and a little bit of time?” Farid says. “Yeah. we’ll figure it out.” But he draws a sharp line between what his team at GetReal can do and what a person doomscrolling on social media is capable of.

Farid’s career began well before the current deepfake boom. He started his academic career at Dartmouth College in 1999, when, as he recalls, the world was largely analog. Film photographs were still common. Digital cameras were emerging. The internet was developing. Social media “didn’t really exist.”.

At the time, Farid says he focused on digital evidence that was “inherently malleable in the courts of law.” In his telling, nobody saw it as a problem at first. Then the digital revolution accelerated, and the issue stopped being niche and started showing up everywhere.

He describes how the field he helped shape began as a “bespoke, niche, tiny, weird field called digital forensics,” driven by him and a group of graduate students at Dartmouth writing papers. The question they kept hearing was blunt: “This is cool, but what does this have to do with anything?”

As citizen journalists emerged and newsrooms began handling user-submitted media, the problem went from occasional to constant. Farid says the work used to involve hearing from media outlets and courts of law once a month and national security once a year. Over the years, it expanded to daily interruptions.

The shift can be seen in how fakes used to be created—and how they’re created now.

In the early days, Farid says he was mostly thinking about photographs because video was much harder to manipulate. Video has 24 to 30 frames a second and an audio track. while images were easier to tamper with using tools like Photoshop. Back then, he says, manipulation still required skill. That meant mistakes—artifacts, misaligned shadows, incorrect geometry, wrong sizes. Sometimes metadata would also reveal that a photo had been edited in Photoshop.

Today, he says, the process has changed so drastically that “you don’t need skill. You don’t need time. You don’t need anything.” All it takes is a keyboard and an internet connection. Farid describes a new reality where someone can type instructions for “this to this image or audio or video. ” and AI can take over. He says the results can be “remarkable things” that were unimaginable five to 10 years ago.

He expects the problem to spread beyond the uncanny valley in waves. Images were probably the first to pass through it. Voice was next, he says, with “the inflection, the laughing, and the pauses.” Video is moving through the uncanny valley now.

Farid draws a practical distinction in length and presentation. If someone shows him a 30-minute HD video, he says it’s probably not AI. But if the clip is 15 to 30 seconds—the typical format people see online—he says it becomes hard to tell from visual cues alone.

He also notes a technical trend: AI-generated videos used to be about four seconds, and now some can reach 30 or 40 seconds by stitching them together. “The content will get better,” Farid says. It will become cheaper, easier to use, and more widespread.

Even so. he argues there’s still a way to investigate fakes—and the science is not just about matching human intuition. Generative AI. he says. “doesn’t know anything about the 3D world.” Farid qualifies that statement with “air quotes.” What matters. in his view. is that AI can still produce outputs with subtly wrong physics.

He says the detection method depends on whether something violates physical plausibility. “As long as you do something that is physically implausible. we have a signal that we can detect.” When the flaws are found. Farid describes it as fast and sometimes relatively easy: once something is clearly wrong. the inquiry can effectively end.

The harder problem is the opposite—authenticating something. Farid says that requires running test after test, and not finding anything wrong. “Does that mean it’s real? Not really,” he says. It only means the team didn’t find an issue.

On average, he says the work can take about an hour. But on the internet. he calls an hour “a long time.” Usually. he says. a call comes in about something. and “there’s already a million views on it.” The team investigates. speaks with a reporter. and the report goes out. In that window, Farid says the content can move from 1 million views to 10 million.

In that sense, he describes his work as “a little bit of a postmortem.” Fact-checking arrives after the attention has already shifted—and the stakes keep climbing.

“I’m not sure how you can have a stable democracy without a shared sense of reality,” Farid says. He doesn’t frame it as a debate about policy. Instead, he describes a broader societal breakdown: arguments are increasingly about whether basic facts even hold.

Farid says people are no longer debating tax rates or the role of government. or even the role of foreign policy—issues that society can disagree on. He says the fight now can resemble a dispute over fundamentals: “We are arguing about whether two plus two is four. I say two plus two is four, and the other person says, ‘No, it’s not. It’s applesauce.’”.

He connects the personal loss of certainty to consequences that extend far beyond social media feeds. Farid says the stakes of being wrong are higher because people are being put in jail, geopolitical decisions are being made, and reporting is being used to inform “eight billion people.”

The fear that drives his warning is not just that fakes are becoming more convincing. It’s that society is losing a shared sense of reality—where the public can accept what happened and where disagreement can still exist without collapsing into, “No, it didn’t.”

deepfakes Hany Farid digital forensics GetReal cybersecurity startup AI-generated video misinformation fact-checking social media democracy media authenticity