I used to start every morning with what I called the presumption of authenticity. If someone showed me a video, I assumed it was real until someone gave me a damn good reason to think otherwise. That was the rule of the road for the last thirty years of digital media. But that world is gone. We have officially crossed the rubicon into the presumption of fakeness.
This isn't just a philosophical shift; it’s a fundamental change in how we have to build and consume technology. Recently, a story broke about an Iranian news agency airing footage of a U.S. Tomahawk missile allegedly hitting a school. The stakes couldn't have been higher—reports of 175 casualties, mostly children. In a pre-AI world, that video would have sparked an immediate international crisis. Today? Everyone’s first reaction, including mine, was to assume it was a deepfake.
The Burden of Proof Has Shifted
For decades, forensic experts spent their time looking for the "gotcha" moment—the digital artifact, the warped pixel, or the physics-defying shadow that proved a video was manipulated. We were the digital detectives. But as the tools of creation (generative AI) have outpaced the tools of detection, the burden of proof has flipped. It is no longer the skeptic’s job to prove a video is fake. It is now the creator’s job to prove it is real.
When I analyzed that Tomahawk video, I had to go back to basics. I aligned frames to stationary landmarks to check the trajectory. I calculated the pixel-to-meter ratio to see if a six-meter missile made sense at that distance. I even timed the delay between the visual impact and the sound. Everything checked out, but even then, I hesitated. Why? Because the absence of a mistake doesn't mean something is real anymore. It just means the AI didn't mess up.
This is a massive problem for builders. If we can't trust the pixels on our screens, the social fabric of the internet starts to unravel. The solution isn't just better forensics; it’s a new architecture for truth.
Provenance Over Detection
If you're building in the AI or media space right now, you need to stop obsessing over "catching the bad guys" and start focusing on "empowering the good guys." Detection is a cat-and-mouse game where the cat is getting old and the mouse is using a jetpack. Instead, we need to lean into provenance.
We are seeing the rise of standards like C2PA—essentially a cryptographic manifest attached to a file at the moment of creation. Think of it like a digital birth certificate. It records the device, the time, the location, and every edit made since the shutter clicked. Apple is already experimenting with this, allowing camera sensors to sign raw pixel data the moment they hit the sensor.
For founders, this is where the opportunity lies. Don't just build another "AI detector" that will be obsolete in six months. Build the infrastructure that allows a journalist, a court, or a regular user to verify that a stream of data actually came from a physical lens in the real world.
The Real-Time Threat
It’s not just static videos we have to worry about. We’re moving into the era of real-time impersonation. We’ve already seen cases where high-level politicians, like Senator Ben Cardin, were targeted in video calls by someone using a face-swap deepfake of a foreign official. When the "Foreign Minister" starts asking weird questions about long-range missiles, you realize you aren't just looking at a video—you're interacting with a weaponized avatar.
Real-time fakes have to run at about 24 frames per second to look natural. To achieve that speed, the software takes shortcuts. It simplifies facial textures and occasionally loses sync between the audio and the lip movement. Humans often miss these micro-glitches because our brains want to believe what we see. But machines don't have to be so gullible.
What This Means for Builders
If you are building communication platforms, you have to assume that every incoming video feed is potentially synthetic. The future of Zoom, Teams, and Google Meet isn't just better filters; it’s real-time verification. Imagine a world where your video calling software checks the cryptographic signature of every frame as it arrives. If the signature doesn't match a verified physical sensor, you get a warning.
Here is the reality for founders: the era of "trust but verify" is dead. We are now in the era of "verify or ignore."
- Stop chasing artifacts: Detection is a losing battle in the long run.
- Adopt C2PA: Make provenance a first-class citizen in your tech stack.
- Hardware matters: The battle for truth will eventually be won or lost at the sensor level.
The goal isn't to build a world where fakes don't exist—that's impossible. The goal is to build a world where the truth is so well-documented that fakes simply don't matter.
We are currently in a period of deep skepticism, and honestly, we should be. It’s the only rational response to the current state of technology. But by shifting our focus from forensic detection to cryptographic provenance, we can eventually get back to a place where we can trust our eyes again. Until then, stay skeptical and keep building the tools that make honesty the default.
Read the original at IEEE Spectrum →