85% Say It's Hard to Tell Scams From the Real Thing, Up From 66% a Year Earlier

On our reading, the most important AI statistic of the week isn't a benchmark score. It's this: 88% of people surveyed say it's becoming harder to tell what's real online.

That's the headline finding of a survey published on 10 June by security firm Malwarebytes, which polled 1,500 adults across the US, UK, Germany, Austria and Switzerland in March. Malwarebytes sells scam-protection software; its report says "Scam Guard is our scam detector". The figure that, on our reading, should worry policymakers most: 85% say it can be hard to tell scams apart from the real thing, up from 66% in Malwarebytes' survey a year earlier. That's a nineteen-point rise, though Malwarebytes' report gives no methodology for the 2025 survey, so how closely the two compare is unclear.

What changed

The survey records what people say, not why. Malwarebytes' report offers its own explanation: "Scammers are exploiting every crack, using AI to further blur the line between real and fake." It adds that image and video manipulation, voice cloning and deepfakes "have made it increasingly difficult for people to spot a scam."

The FBI's 2025 Internet Crime Report makes a related point: "AI-enabled synthetic content is becoming increasingly difficult to detect and easier to make". Michael Machtinger, deputy assistant director of the FBI Cyber Division, told the Wall Street Journal: "AI-created fraudulent communications can look very official and very legitimate to even the most trained individuals." On our reading, what has changed is the tools, not the people.

The costs are already real. According to Malwarebytes' summary of the FBI's 2025 Internet Crime Report, 22,364 complaints to the FBI in 2025 reported $893 million (nearly $900 million) in AI-related scam losses, and those figures represent only the reported losses.

Why this is a policy story, not a tech story

The instinct is to treat this as a consumer-awareness problem: teach people the tells. On our reading, the survey points the other way. If 85% of those surveyed already say it can be hard to tell a scam from the real thing, spot-the-fake education on its own looks close to its limits.

That, on our reading, shifts the burden to two places:

Provenance infrastructure. We think surveys like this one strengthen the case for content credentials (the Coalition for Content Provenance and Authenticity, C2PA, describes its "open technical standard" as a way to "establish the origin and edits of digital content"), to be visible by default rather than buried in metadata.

Liability and regulation. If people find it hard to tell real from fake, the legal question, as we see it, becomes who pays when they're deceived: the platform that delivered the deepfake, the AI company whose model made it, or no one. We expect numbers like Malwarebytes' to be cited in that debate.

The honest takeaway

There's a version of this story that's pure alarm, and on our reading it should be resisted: this survey measures how hard people say it is to tell, not tested ability, and "can't tell" headlines, ours included, often report self-reported difficulty as if it were inability. People may be better at spotting fakes than they believe, or worse.

But on our reading the direction of travel points one way: authenticity is becoming something that has to be proven, not assumed.

Correction, 3 October 2026: An earlier version of this article said 85% of people "can't tell" scams from reality. Malwarebytes' survey of 1,500 adults in five countries found that 85% said it can be hard to tell a scam from the real thing, up from 66% in the company's 2025 survey; the report gives no methodology for the 2025 survey, so how closely the two compare is unclear. The headline, standfirst, text and audio narration have been corrected. We also removed a statement that the separate 88% figure was rising, because the survey gives no earlier comparison for it, and specified that the FBI's figure covers losses reported in 2025. We removed claims that our cited sources did not support, including a German court ruling against Google, a draft US federal AI framework, scam operations hiring AI models, a 15,500-domain scam network, specific advances in AI tools, the state of content-credential deployment and the general safety of online content and transactions, and replaced the unsupported account of tool advances with what Malwarebytes and the FBI said. We removed a 98% figure that appears in no source and labelled our own analysis as such. The survey was published on 10 June, not on the day of our article. We also replaced two secondary sources (Help Net Security and Security Boulevard) with Malwarebytes' full report and its original post on the FBI figures, added the FBI's 2025 report and the C2PA's own site as sources, and noted that Malwarebytes sells scam-protection software.