Deepfakes Scare People

Executive summary

YouTube's decision to open its AI-powered Likeness Detection tool to every adult user is landing at exactly the right moment — and most people know it. A new pulse survey of 188 adults finds that 76.5% feel positively about the face-scan deepfake detection feature, with nearly half (44.1%) calling it "much-needed protection."

The enthusiasm is well-calibrated. Resemble AI's 2025 Threat Report documents 1,567 verified deepfake incidents generating 296.4 billion combined media impressions and over $1.28 billion in fraud losses — and federal legislation to protect victims has been stalled in the House Judiciary Committee for ten months. For millions of ordinary users, YouTube's tool is now one of the only scalable options available.

But support is not the same as trust. Free-response answers reveal a polarized split: respondents lean toward privacy over convenience (mean score −0.37 on a −1 to +1 scale), and concern about biometric data misuse, data resale, and function creep runs through nearly every skeptical response. Getting people to actually enroll will require more than a good idea — it will require a credible data-protection promise.

Meanwhile, deepfake worry is widespread even among users who have never seen one, signaling that fear is driven by awareness and perceived risk, not personal experience.

Context

YouTube quietly rolled out Likeness Detection — a face-scanning AI system that flags unauthorized deepfake videos and lets users request removal — to all adults 18 and older in May 2026. The expansion is significant: previously limited to verified creators and talent agencies, the tool now gives any adult the ability to continuously monitor YouTube for content that mimics their face.

The mechanics work similarly to Content ID. Users enroll by submitting a brief selfie video and a government-issued ID. YouTube's system then performs a one-time scan of newly uploaded videos, comparing them against the enrolled likeness template. If a match is detected, users can request removal under YouTube's privacy policy. Critically, non-enrolled individuals' scan data is deleted immediately after matching — but for enrolled users, likeness templates and identity information are stored for up to three years from the last login or until users withdraw consent.

The timing of this expansion reflects a deepening crisis. Resemble AI's 2025 Threat Report tracked 1,567 verified deepfake incidents in a single year, with zero lag between major world events and the first deepfake responses — deployment within hours became standard. Twenty percent of those incidents involved non-consensual intimate imagery (NCII) or child sexual abuse material. The Alan Turing Institute finds 90.4% of the general public is already concerned about deepfakes, with women disproportionately targeted.

Against this backdrop, the pulse survey gathered 188 adult responses across four questions — two multiple-choice and two open-ended — probing attitudes toward the tool, concerns about biometric data, personal worry about being deepfaked, and prior exposure to deepfake content. The results offer a timely read on public sentiment as YouTube's enrollment window opens to mainstream users for the first time.

Findings

Most users see the tool as protection — but a vocal minority won't touch it

The headline number is unambiguous: 76.5% of respondents feel positively about YouTube's face-scan feature, with 44.1% choosing "very positive — this is much-needed protection." That is not passive approval; it reflects an audience that has registered the scale of the deepfake threat and views platform-level intervention as legitimate.

But 10.6% hold negative views, and they are not a fringe. Their concerns — surfaced in open-ended responses — center on a specific and well-documented fear: that a face scan submitted to detect deepfakes could itself become a vulnerability. Respondents flagged worries about biometric data being sold to third parties, used to train AI models, accessed in a breach, or repurposed for law enforcement purposes they never consented to. This is textbook function creep anxiety, and it is rational: only 15.4% of people in external surveys trust private companies with their facial data, and nearly half have opted out of services entirely over privacy concerns.

The free-response dimension captures the split precisely. On a scale of −1 (pure privacy concern) to +1 (pure convenience acceptance), respondents land at a mean of −0.37 — modestly privacy-leaning, but with a polarized distribution: many people cluster near both ends rather than converging in the middle. The 11.7% who feel neutral represent a reachable middle — users who aren't opposed but haven't yet been given a compelling reason to enroll.

Fear of being deepfaked doesn't require ever seeing one

Just over half of respondents — 51% — have encountered at least one deepfake video online, with 30.3% saying they've seen several and 20.7% saying they've seen one or two. That leaves 49% who have had no direct exposure: 27.7% who know what deepfakes are but haven't seen one, and 21.3% who aren't even sure what the term means.

The striking finding is what happens in the unexposed group. Respondents who expressed the most worry about someone creating fake videos of their face were disproportionately likely to have never encountered a deepfake — and in some cases to be unfamiliar with the concept entirely. Worry, in other words, is decoupled from exposure. It is driven by awareness of possibility and perceived personal risk, not by firsthand experience of the threat.

This has a direct implication for YouTube's enrollment numbers. If latent concern already exists among people who haven't seen a deepfake, education campaigns that make the threat concrete — and link it clearly to a simple enrollment process — could convert that ambient anxiety into active protection-seeking behavior.

Who worries most — and who embraces the tool — is predictable from personality

Trait analysis across the respondent sample reveals consistent psychological patterns beneath the headline numbers.

Higher Agreeableness (the personality trait associated with empathy, cooperation, and sensitivity to others' harm) predicts greater worry about being deepfaked (r=0.243, p=0.003). This is counterintuitive at first glance — a Nature study on general internet privacy finds that agreeableness actually reduces privacy concern. The difference may be that deepfake worry taps into harm-sensitivity and empathy rather than general data vigilance: agreeable people are imagining what it would feel like to be victimized, not abstractly calculating surveillance risk.

On the other side of the ledger, higher Extraversion (r=0.230) and higher Prism Influence scores (r=0.212) both predict more positive attitudes toward the YouTube feature. Socially engaged, outward-facing users — people who are already managing a public presence — see the most personal relevance in likeness protection. For them, the tool solves a real and immediate problem.

Conversely, respondents high in Persistence and Meticulousness report less worry about deepfakes (r=−0.198 and r=−0.193, respectively). These traits tend to correlate with systematic, evidence-based risk assessment — suggesting that highly methodical thinkers are discounting the personal probability of being targeted, even as they may intellectually accept the threat exists at scale.

The practical read: YouTube's messaging may need to be segmented. For empathetic, socially active users, leading with personal protection and harm prevention will resonate. For analytical skeptics, the more effective pitch may be accuracy, data minimization, and the opt-out guarantee.

Conclusion

YouTube's Likeness Detection expansion has broad public backing — but the real test is enrollment, not sentiment. The 76.5% who feel positively about the tool have not yet submitted a government ID and a face scan. The gap between approving of an idea and trusting a platform with biometric data is exactly where this initiative could stall.

The federal legislative void makes that trust gap more consequential, not less. With the Defiance Act stranded in the House Judiciary Committee and state laws like Michigan's still too new to show deterrence effects, platform tools are carrying policy weight they were never designed to bear. If YouTube's removal request volume stays "very small" — the company's own description — the question isn't whether the feature works technically, but whether users will engage with it at scale.

Three things to watch: whether YouTube publishes transparent data on template storage, deletion rates, and removal outcomes; whether awareness campaigns close the knowledge gap among the 21.3% who don't yet know what deepfakes are; and whether the Defiance Act moves in the House before the next election cycle, when deepfake deployment — already measured in hours after a major news event — is near-certain to accelerate.