The fluent cohort is the scammed cohort: 70% of 18-to-22s hit by an AI scam

Malwarebytes has published survey figures putting 70 percent of 18-to-22-year-olds as having experienced an AI-related scam in the past year, compared with half of the general population ^[1]. That is a 20 point gap, roughly 1.4 times the rate ^[2], and it runs against any awareness program that sorts risk by assumed digital fluency.

The company's own reading is exposure rather than carelessness. Malwarebytes argues that the platforms young people depend on for daily life, social feeds where manipulated and real content sit together, marketplaces full of fake storefronts and reviews, and messaging channels where threats can be personalised, now double as entry points for AI-enabled attacks ^[3]. It describes the result as a new safety burden: use AI, judge its output, protect your identity, and dodge increasingly personalised scams, all at once ^[4].

The scam-type breakdown is where the argument gets sharper. On deepfake or virtual kidnapping scams, 19 percent of young people report being victims versus 8 percent of the general population ^[5], and 43 percent report being targeted versus 26 percent ^[6]. That is about 2.4 times the victimisation rate on about 1.65 times the targeting rate ^[7][8]. Treating victims as a subset of targets, which the published post does not confirm the questions were structured to do ^[9], implies roughly 44 percent of targeted young people were victimised against roughly 31 percent of the general population ^[10].

Compare sextortion, where the pattern flips. Young people are targeted far more often, 38 percent versus 24 percent ^[11], yet victimisation is almost identical, 8 percent versus 7 percent ^[12], implying a lower conversion rate for the younger group, roughly 21 percent versus 29 percent ^[13]. Romance scams behave similarly: 46 percent targeted versus 33 percent ^[14], 12 percent victimised versus 10 percent ^[15], again implying slightly lower conversion ^[16]. Extortion overall runs 56 percent targeted versus 42 percent ^[17] and 24 percent victimised versus 17 percent ^[18]; impersonation, 47 versus 35 targeted ^[19] and 14 versus 10 victimised ^[20].

Read together, fluency appears to hold up reasonably well against the old emotional plays and to fail specifically against synthetic media. The trust data is consistent with that. Malwarebytes reports 44 percent of young people have seen AI give information they knew or later learned was wrong, against 30 percent of the general population ^[21]; 23 percent report negative consequences from AI advice versus 16 percent ^[22]; 18 percent report emotional harm versus 12 percent ^[23]. Half of 18-to-22-year-olds strongly agree it is getting harder to tell whether content is genuinely human ^[24].

Behaviour has already moved. Forty-seven percent say AI changed how much they trust reviews or content, against 37 percent ^[25]; 35 percent changed how they shop, against 26 percent ^[26]; 32 percent changed how they present themselves professionally, against 20 percent ^[27]; 19 percent changed how they date, against 10 percent ^[28]. One respondent told Malwarebytes that dating online is no longer an option because "you just never know what is or isn't AI" ^[29].

Two things to watch. First, the methodology: the post as published does not state sample size, fielding dates, or how respondents were recruited ^[9], so treat the point gaps as directional. Second, whether the deepfake conversion gap shows up in incident data rather than self-report. If it does, the training population that needs synthetic-media verification drills first is the one most programs assume already knows.