Why is it getting harder to spot a deepfake — and what should you do when you can’t trust your own eyes?

📺🤖 A fraudster joins a video call. Within seconds, real-time deepfake software replaces their face with someone else’s — a boss, a banker, a government official — live, without the illusion breaking. Tools like Haotian AI, a piece of Chinese-developed software marketed to scammers, can already do exactly this on WhatsApp, Zoom, and Teams, according to 404 Media. This isn’t a future risk. It’s happening now — and the human eye can no longer reliably catch it. Even the world’s leading expert in digital forensics says he’s started failing his own tests.

A little background: Deepfakes are AI-generated or manipulated images, videos, or audio recordings that make it appear like someone did or said something they never did. When they first emerged, telltale signs were easy to spot: unnatural blinking, distorted facial features, robotic movement. Modern AI has largely eliminated those artifacts. The flaws people have learned to look for no longer exist at scale.

You probably can’t spot one — and the data proves it

The world’s leading digital forensics expert, Hany Farid, told The New York Times that over the past six months, he has stopped trusting his own eyes. If the field’s foremost expert is struggling, the rest of us don’t stand much of a chance. When researchers showed US respondents 16 visuals — eight real, eight AI-generated — they averaged a detection score of just 0.07 on a scale of -1 to 1, where 0 represents pure chance, according to Veriff’s Deepfakes Report 2026. The result is statistically indistinguishable from a coin flip. Half of those respondents described themselves as confident in their ability to detect manipulated media. That confidence bore almost no relationship to their actual performance.

Video content proved especially deceptive: in one side-by-side test, 70% of respondents misidentified the AI-generated video as real, according to the same report. The tactics most people rely on to spot fakes — unnatural skin texture, odd facial proportions, jerky movement — are precisely the artifacts that modern AI tools are engineered to eliminate, Veriff found. People are looking for flaws that no longer exist.

SOUND SMART- A detection score of 0.07 on a -1 to 1 scale means respondents performed almost identically to random guessing. A score of 1 would mean perfect detection; a score of -1 would mean getting it wrong every time.

What deepfakes are already being used for

Recent incidents: The Bank of England was forced to issue a public statement after a deepfake video depicting governor Andrew Bailey in a physical altercation with politician Nigel Farage circulated widely — and a significant share of the public failed to identify it as fake, according to The Guardian. In Canada, a man is facing 79 charges after creating dozens of deepfakes of women in violent and sexual positions without their consent, according to CBC.

The security stakes go further. Shortly after Russia began its invasion of Ukraine, a deepfake video of President Volodymyr Zelenskyy apparently urging the military to lay down their weapons and surrender spread on social media — one of the first high-profile examples of a deepfake being weaponized in active armed conflict, according to Northwestern University’s Buffett Institute (pdf). Researchers have since identified a range of potential security applications for the technology, including falsifying military orders, discrediting political leaders, and exploiting national tensions to deepen polarization, the institute notes.

We’re all prone to it

Some 79% of US respondents said they are concerned about deepfake-driven personal fraud — the top fear in the survey — though Americans are also more likely than their UK and Brazilian counterparts to trust platforms to manage AI-generated content on their behalf, according to Veriff. That combination of high concern and outsourced responsibility is the condition fraud exploits most effectively.

Roughly 7% of respondents fall into a high-risk category: poor detection performance, high confidence, and a habit of rarely verifying suspicious content, making them a persistent and predictable target for deepfake-driven fraud, according to the same report. The implication, per Farid, is a shift in how people should approach any media involving their finances or personal information: away from visual confirmation entirely, and toward verification through trusted, independent sources.

Farid’s own advice, shared with The New York Times, points in one direction: stop relying on your eyes entirely. For any media involving financial or personal information, verify through trusted, independent sources — not through visual inspection alone.

What Egypt’s framework looks like — and where it stops

Egypt doesn’t yet face the same volume of deepfake incidents as the US, UK, or Canada — but the trajectory is clear, and policymakers are moving. Egypt’s National Guidelines for Generative AI emphasize transparency and mandatory disclosure of AI-generated content, with heightened safeguards recommended for high-risk sectors like government. Dar Al Ifta prohibited the creation of deepfakes in 2022, citing their potential for fraud, defamation, and the spread of misinformation.

Policy is a floor, not a ceiling. Organizations should move verification systems beyond traditional means and build infrastructure capable of identifying AI-generated content before it causes damage, according to the World Economic Forum — a standard that applies as much to Egyptian banks, media organizations, and government bodies as it does to their counterparts elsewhere.