Skip to main content

Facebook develops new method to reverse-engineer deepfakes and track their source

Facebook develops new method to reverse-engineer deepfakes and track their source

/

The work could help future deepfake investigations

Share this story

Deepfakes aren’t a big problem on Facebook right now, but the company continues to fund research into the technology to guard against future threats. Its latest work is a collaboration with academics from Michigan State University (MSU), with the combined team creating a method to reverse-engineer deepfakes: analyzing AI-generated imagery to reveal identifying characteristics of the machine learning model that created it.

The work is useful as it could help Facebook track down bad actors spreading deepfakes on its various social networks. This content might include misinformation but also non-consensual pornography — a depressingly common application of deepfake technology. Right now, the work is still in the research stage and isn’t ready to be deployed.

The method could help track down those spreading deepfakes online

Previous studies in this area have been able to determine which known AI model generated a deepfake, but this work, led by MSU’s Vishal Asnani, goes a step further by identifying the architectural traits of unknown models. These traits, known as hyperparameters, have to be tuned in each machine learning model like parts in an engine. Collectively, they leave a unique fingerprint on the finished image that can then be used to identify its source.

Identifying the traits of unknown models is important, Facebook research lead Tal Hassner tells The Verge, because deepfake software is extremely easy to customize. This potentially allows bad actors to cover their tracks if investigators were trying to trace their activity.

Examples of deepfakes include these fake faces, generated by a well-known AI model called StyleGAN.
Examples of deepfakes include these fake faces, generated by a well-known AI model called StyleGAN.
Image: The Verge

“Let’s assume a bad actor is generating lots of different deepfakes and uploads them on different platforms to different users,” says Hassner. “If this is a new AI model nobody’s seen before, then there’s very little that we could have said about it in the past. Now, we’re able to say, ‘Look, the picture that was uploaded here, the picture that was uploaded there, all of them came from the same model.’ And if we were able to seize the laptop or computer [used to generate the content], we will be able to say, ‘This is the culprit.’”

Hassner compares the work to forensic techniques used to identify which model of camera was used to take a picture by looking for patterns in the resulting image. “Not everybody can create their own camera, though,” he says. “Whereas anyone with a reasonable amount of experience and standard computer can cook their own model that generates deepfakes.”

Not only can the resulting algorithm fingerprint the traits of a generative model, but it can also identify which known model created an image and whether an image is a deepfake in the first place. “On standard benchmarks, we get state-of-the-art results,” says Hassner.

Deepfake detection is still an “unsolved problem”

But it’s important to note that even these state-of-the-art results are far from reliable. When Facebook held a deepfake detection competition last year, the winning algorithm was only able to detect AI-manipulated videos 65.18 percent of the time. Researchers involved said that spotting deepfakes using algorithms is still very much an “unsolved problem.”

Part of the reason for this is that the field of generative AI is extremely active. New techniques are published every day, and it’s nearly impossible for any filter to keep up.

Those involved in the field are keenly aware of this dynamic, and when asked if publishing this new fingerprinting algorithm will lead to research that can go undetected by these methods, Hassner agrees. “I would expect so,” he says. “This is a cat and mouse game, and it continues to be a cat and mouse game.”

Today’s Storystream

Feed refreshed 12:00 AM UTC Dimorphos didn’t even see it coming

R
Twitter
Richard Lawler12:00 AM UTC
A direct strike at 14,000 mph.

The Double Asteroid Redirection Test (DART) scored a hit on the asteroid Dimorphos, but as Mary Beth Griggs explains, the real science work is just beginning.

Now planetary scientists will wait to see how the impact changed the asteroid’s orbit, and to download pictures from DART’s LICIACube satellite which had a front-row seat to the crash.


M
The Verge
We’re about an hour away from a space crash.

At 7:14PM ET, a NASA spacecraft is going to smash into an asteroid! Coverage of the collision — called the Double Asteroid Redirection Test — is now live.


E
Twitter
Emma RothSep 26
There’s a surprise in the sky tonight.

Jupiter will be about 367 million miles away from Earth this evening. While that may seem like a long way, it’s the closest it’s been to our home planet since 1963.

During this time, Jupiter will be visible to the naked eye (but binoculars can help). You can check where and when you can get a glimpse of the gas giant from this website.


Asian America learns how to hit back

The desperate, confused, righteous campaign to stop Asian hate

Esther WangSep 26
E
Twitter
Emma RothSep 26
Missing classic Mario?

One fan, who goes by the name Metroid Mike 64 on Twitter, just built a full-on 2D Mario game inside Super Mario Maker 2 complete with 40 levels and eight worlds.

Looking at the gameplay shared on Twitter is enough to make me want to break out my SNES, or at least buy Super Mario Maker 2 so I can play this epic retro revamp.


R
External Link
Russell BrandomSep 26
The US might still force TikTok into a data security deal with Oracle.

The New York Times says the White House is still working on TikTok’s Trump-era data security deal, which has been in a weird limbo for nearly two years now. The terms are basically the same: Oracle plays babysitter but the app doesn’t get banned. Maybe it will happen now, though?


R
Youtube
Richard LawlerSep 26
Don’t miss this dive into Guillermo del Toro’s stop-motion Pinocchio flick.

Andrew Webster and Charles Pulliam-Moore covered Netflix’s Tudum reveals (yes, it’s going to keep using that brand name) over the weekend as the streamer showed off things that haven’t been canceled yet.

Beyond The Way of the Househusband season two news and timing information about two The Witcher projects, you should make time for this incredible behind-the-scenes video showing the process of making Pinocchio.


R
External Link
Russell BrandomSep 26
Edward Snowden has been granted Russian citizenship.

The NSA whistleblower has been living in Russia for the 9 years — first as a refugee, then on a series of temporary residency permits. He applied for Russian citizenship in November 2020, but has said he won’t renounce his status as a U.S. citizen.


E
External Link
Emma RothSep 26
Netflix’s gaming bet gets even bigger.

Even though fewer than one percent of Netflix subscribers have tried its mobile games, Netflix just opened up another studio in Finland after acquiring the Helsinki-based Next Games earlier this year.

The former vice president of Zynga Games, Marko Lastikka, will serve as the studio director. His track record includes working on SimCity BuildIt for EA and FarmVille 3.


A
External Link
Vietnam’s EV aspirant is giving big Potemkin village vibes

Idle equipment, absent workers, deserted villages, an empty swimming pool. VinFast is Vietnam’s answer to Tesla, with the goal of making 1 million EVs in the next 5-6 years to sell to customers US, Canada and Europe. With these lofty goals, the company invited a bunch of social media influencers, as well as some auto journalists, on a “a four-day, multicity extravaganza” that seemed more weird than convincing, according to Bloomberg.


J
James VincentSep 26
Today, 39 years ago, the world didn’t end.

And it’s thanks to one man: Stanislav Petrov, a USSR military officer who, on September 26th, 1983, took the decision not to launch a retaliatory nuclear attack against the US. Petrov correctly guessed that satellite readings showing inbound nukes were faulty, and so likely saved the world from nuclear war. As journalist Tom Chivers put it on Twitter, “Happy Stanislav Petrov Day to those who celebrate!” Read more about Petrov’s life here.


Soviet Colonel who prevented 1983 nuclear response
Photo by Scott Peterson/Getty Images
J
The Verge
James VincentSep 26
Deepfakes were made for Disney.

You might have seen the news this weekend that the voice of James Earl Jones is being cloned using AI so his performance as Darth Vader in Star Wars can live on forever.

Reading the story, it struck me how perfect deepfakes are for Disney — a company that profits from original characters, fans' nostalgia, and an uncanny ability to twist copyright law to its liking. And now, with deepfakes, Disney’s most iconic performances will live on forever, ensuring the magic never dies.


E
External Link
Hurricane Fiona ratcheted up tensions about crypto bros in Puerto Rico.

“An official emergency has been declared, which means in the tax program, your physical presence time is suspended,” a crypto investor posted on TikTok. “So I am headed out of the island.” Perhaps predictably, locals are furious.


R
The Verge
Richard LawlerSep 26
Teen hacking suspect linked to GTA 6 leak and Uber security breach charged in London.

City of London police tweeted Saturday that the teenager arrested on suspicion of hacking has been charged with “two counts of breach of bail conditions and two counts of computer misuse.”

They haven’t confirmed any connection with the GTA 6 leak or Uber hack, but the details line up with those incidents, as well as a suspect arrested this spring for the Lapsus$ breaches.