I was reading a crazy story recently in the New York Times about the world’s foremost deepfake expert, Hany Farid. You may not recognize the name, but he’s the guy who can spot a fake Internet video or image better than anyone else.

Or at least he once could. Apparently, he can’t tell anymore.

That is, with the technology advancements, he’s now struggling to identify what’s real from what’s created by AI.

That should be pretty sobering for all of us ordinary people, who don’t have the training, the technology, the skills, or the eye for detail that the world’s greatest deepfake expert possesses.

And it raises an uncomfortable question: What happens in that kind of world? Not just to marketing, but almost everything we see?

Why Experts Are Struggling Now

Early deepfakes and AI images often had obvious glitches (hands, eyes, teeth, lighting inconsistencies), and they lacked low-level “statistical signatures” of real photographs, such as sensor noise and lens artifacts.

Modern generators explicitly learn and reproduce those low-level patterns and can add realistic noise and imperfections, so many of the pixel-level forensic cues Farid originally used no longer work.

Because most current systems generate media from scratch rather than editing an existing file, several classic manipulation-detection methods (which looked for inconsistencies introduced by Photoshop-style editing) are less effective or fail entirely.

The Implications of Not Being Able to Tell What’s Real

In a world where no one can reliably tell real from fake, you don’t just get more scams. What I fear is that you also get a slow collapse of the basic trust scaffolding that modern societies and markets rely on. The biggest risk isn’t only “people believe lies,” it’s also “people stop believing anything,” which is arguably worse.

We’re Living in the ‘Age of Impersonation’

Hany Farid and others increasingly describe this as an “age of impersonation,” where anyone can make anyone appear to say or do anything in high fidelity. That possibility redefines what identity means online: your voice, face, and writing style become parameters others can synthesize, not unique proofs that “this is really you.”

Also, just think about the phrase: I saw/heard it myself (with my own eyes/ears).”

Will that even work anymore?

What gets put in its place may be something more like “forensic skepticism”: only content whose origin and chain of custody can be demonstrated earns belief. But as demonstrated above, even that can be a challenge.

That’s on a personal level too.

At a societal scale, who knows what happens? After all, most people and most institutions are not set up to run evidence authentication as a daily habit.

Cybercrime, Identity Theft, Fraud, and the Erosion of Economic Trust

Deepfakes and synthetic media are already a huge threat. When criminals can generate real‑time cloned voices, convincing video calls, and hyper‑personalized messages using leaked data, the cost of running sophisticated scams drops close to zero. Whereas traditional identity theft was mostly about numbers and static secrets, deepfakes extend that into behavioral and biometric impersonation: face, voice, gait, even keystroke patterns can be copied or approximated. The economic consequences of this could be massive:

• • Contracts and agreements: If a “video of the CEO authorizing this transfer” can be fabricated, organizations need new internal controls, which add friction and cost to everyday operations.

• • Brand and financial reputation: Synthetic announcements, forged earnings calls, or fake crisis videos can move markets or damage brands before corrections catch up.
  • Consumer trust in digital channels: If customers repeatedly encounter convincing fraud posing as banks, retailers, or health systems, the default may become distrust of all digital communication, which harms legitimate businesses too.

• • The harm becomes reputational and relational, not only financial: Someone can place you at a protest, in a compromising sexual situation, making racist comments, or “admitting” to crimes, in ways that can damage relationships, employment, and even physical safety.

Elections, Propaganda, and Democracy: A Fragile Democracy Gets Even More So

Deepfake elections research is pretty clear: high‑quality synthetic political content can plausibly swing perceptions, suppress votes, and discredit opponents. And, worse, it weaponizes plausible deniability. Here in the U.S., the introduction of more deepfakes is not just concerning (at a time when a large chunk of its citizens doesn’t know what’s real or true to start with), it’s potentially destabilizing.    

Consider these 3 scenarios:

• • Fabricated scandals: A forged video of a candidate making inflammatory remarks circulates days before an election; fact‑checking lags behind the emotional impact.
  • Deniable truth: A real recording of wrongdoing emerges, and the accused simply labels it a deepfake; supporters, already primed to distrust media, accept that explanation.
  • Flooding: Rather than one decisive fake, voters are bombarded with synthetic clips, doctored “leaks,” and AI‑generated commentary until distinguishing real from fake feels impossible, encouraging apathy or withdrawal.

Social Cohesion and Interpersonal Trust Will Almost Certainly Suffer

At the interpersonal level, synthetic media becomes a new tool for harassment, extortion, and social destruction. Even Farid had his number spoofed, and attackers used AI to clone his voice. Today, he and his wife use a “safe” word to confirm they are each real at the start of any phone call. (It should say something that the world’s foremost deepfake expert and his wife need a safe word.)

But think about it. Most social transactions rely on that inherent trust – the idea that you can tell the real person from a fake one.

Moreover, it goes beyond simple spoofing. What about non‑consensual deepfake pornography? Fake admissions of infidelity or betrayal? Fabricated racist rants, or “leaked” private conversations? All of those can be targeted precisely at an individual’s social network. This could:

• • Increase so-called “ambient” paranoia: People become more cautious about being filmed, about what they share, and about what they believe about others.
  • Fracture communities: Well‑timed fakes can be used to split activist groups, religious communities, companies, or friend networks.
  • Normalize disengagement: If “it could all be fake,” some people retreat from civic life and public discussion entirely, which in turn strengthens more tightly knit, sometimes more extreme, in‑groups.

Law, Governance, and Power: How Do We Decide What’s True?

Legally, societies face messy questions: what counts as defamation in an age where “I never said that” can be both true and a convenient lie? How do courts treat video evidence when both real and fake are plausible? Consider the following implications:

• • New evidentiary standards: Courts relying more on cryptographically signed evidence, device logs, and expert testimony about authenticity, and less on raw audio/video as inherently persuasive.
  • Deepfake‑specific regulations: Some jurisdictions are already experimenting with laws on political deepfakes near elections, non‑consensual sexual deepfakes, and disclosure requirements for synthetic media.
  • Centralization vs pluralism: There’s a real risk that “only a small number of state‑approved or big‑platform verification authorities” becomes the de facto arbiter of reality, which carries its own abuse and capture risks.

Philosophically, this reopens old debates about who gets to define truth in public spaces: decentralized peer structures, professional intermediaries (press, academia), or state and corporate gatekeepers.

We’ve Said It Before: Why Not Pause This Train Before It Goes Off the Cliff?

The sad part is that the above only scratches the surface in terms of the damage this particular AI technology could bring upon society. And there is no scenario in which harm won’t be done to innocent people if we continue to let companies and people use this AI freely.

In a previous blog, we asked the question: With technology, when is enough enough?

After all, no one asked for deepfakes on their Christmas wish lists.[1] Yet, here we are.

Even Dario Amodei, the CEO of Anthropic, one of the largest AI companies in the world, is begging for his industry to be regulated. He clearly sees the risks of not just deepfakes, but the whole thing.  

So, why don’t we do something?

It seems like here in the U.S., in particular, we just want to continue “business as usual.” We either don’t have the time or the inclination to worry about things. We have our “faith” that things will work out because, well, they always have, right?

We also have a Congress that seems devoted to this idea of limited regulation as if it were our religion. The big guys, who bankroll Congressional campaigns, all make their money in the short term. Forget about what happens to the little people. After all, in a world of deepfakes, the attitude must be: “we can tell them anything we want, and they’ll believe it!”    

The sad part is that the time to act is now if we’re to make any inroads at all into these issues. Because if we don’t do something soon, it won’t be just a “safe word” that we’ll all need.

[1] Even if a deepfake is an unintentional consequence, it’s still a consequence.