Deepfake prism refracts fraud landscape ‘supercharged’ by generative AI

Is the avalanche of deepfakes triggering a digital trust apocalypse? The seemingly dire question of deepfake fraud has entered the Prism Project’s refracting lens, as showcased in a recent webinar hosted by Peak IDV.

Mirroring its flagship Biometric Digital Identity Prism Landscape, the Prism Project – a new curator of market landscape frameworks – has crafted a prismatic scheme in which to situate the growing ecosystem of providers tackling the challenge of deepfake media and synthetic identity.

Host C. Maxine Most, a founder of the Prism Project, explains that its prism has “beams” – i.e. market segments – and that the idea for the webinar is to cross beams, so to speak, in order to get a better picture of what’s happening in every corner of the deepfake landscape.

Sponsor guests represent AuthenticID, Keyless and DuckDuckGoose – situated, respectively, in the identity proofing and verification, passwordless authentication and core identity technology beams.

Fielding the first question in the rapid-fire format (branded as a “three-beam stream”), Parya Lofti, CEO of DuckDuckGoose, says that “when anyone, basically, can create a hyper realistic picture of anyone else, it feels like digital apocalypse.” She raises a somewhat harrowing thought: when your face is everywhere – on social media or webinars – and everyone has a way to capture it, the question of ownership is thrown into doubt. She repeats the query of a colleague: “Do you even think that the face you have only belongs to you?”

Blair Cohen is the president and chief evangelist of AuthenticID, and his message is equally troubling: the digital trust apocalypse is already here. “The tools that I saw three months ago, you used to be able to beat those things on video – you could have somebody hold their hand up to their face and these occlusions would appear, there were all kinds of artifacts. I tried two new tools last week, and I found none of that.”

Cohen cites an article reporting on a job for which 71 percent of applicants turned out to be fake people. “You can’t trust Zoom anymore,” he says. “I think we’re going to have to go back to in-person interviewing.”

Thankfully, Tobin Broadfoot, product director at Keyless, is more upbeat. While he concedes that things are getting weirder by the week, he pish-toshes the idea that digital trust is in the midst of its armageddon. “How many millions of transactions took place today, how many millions of identity verifications? I’m still transacting on my phone. We’re okay from that perspective.”

But are we? Broadfoot acknowledges that it remains hard work to keep pace with deepfake fraud. “Now what we have to do when it comes to face ownership, it’s not just how your face looks. It’s also the inherent liveness detection. It’s what your face looks like, plus.”

As Most notes, the narrative about a culture subsumed in deepfakes is pretty convenient for deepfake detection and fraud prevention providers. But as Cohen notes, there is data to back up the claims of a dramatic escalation. He notes a statement that came from the Better Identity Coalition, claiming that deepfake attacks had increased more than 4000 percent year over year.

“Our own numbers are a little bit lower,” Cohen says. “We’re right at about 1700 percent. So I don’t think it’s overhyped.” He points to injection attacks as an active threat that has yet to be solved entirely – the next frontier in deepfake detection.

Algorithms not the only tool for fighting deepfakes – but a necessary one

The proverbial question arises: can we use AI to fight AI?

Broadfoot notes that while AI’s a hugely broad term, “any vendor in this industry that’s still surviving is collecting lots and lots of data and using machine learning, and iterating and iterating and iterating again to make it better and better.” That said, “no one in their right mind is suggesting AI is the only thing you should use to fight against AI deepfakes. It’s one tool in the toolbox.”

Cohen is more blunt: “It’s really the only way that we’re going to be able to detect these anomalies.” The biometrics industry has been using what’s now frequently dubbed “AI” for years, refining the tech to limit bias and improve accuracy. His present concern is the potential “handcuffs” of AI regulation, which he fears will stymie attempts to stay responsive to evolutions in fraud attacks.

Regulation, however, is likely also the only way to establish a foundation of digital trust that can be defended against the wildlands of AI fraud. Sopping up the rancid soup of fraud – selfie verification fraud, deepfake CEO injection attacks, synthetic job applicants, rogue AI agents – will require legislation, innovation and standardization baked into a complex loaf that balances consistency with zest (i.e. privacy with usability).

Maxine Most underlines the need for the whole sector to clearly define its problems, and know what’s available to help solve them. “You have to really be thinking about, what are the problems you‘re trying to solve and how can different technologies across this ecosystem help you build a solution that’s actually going to be useful for your organization?”

Article Topics

AI fraud  |  AuthenticID  |  biometrics  |  deepfake detection  |  deepfakes  |  digital trust  |  DuckDuckGoose  |  generative AI  |  Keyless  |  Maxine Most  |  Peak IDV  |  Prism Project