Avast warns of AI confidence trap over chatbot data

Avast has published UK research suggesting that many adults who believe they can spot AI fakes still share sensitive information with chatbots. The findings point to what it describes as an AI confidence trap.

The study found that 70% of UK adults believe they would recognise an AI fake on sight, yet 45% have shared, or considered sharing, at least one sensitive item with an AI tool. Among the examples, 23% said they had shared, or considered sharing, a CV with an AI chatbot, while 19% said the same about a work document.

The figures suggest a gap between people's confidence in identifying manipulated content and their judgement about what information is safe to give to AI systems. Avast, part of Gen, said that distinction matters because data entered into chatbot tools may be stored beyond the immediate conversation.

Richard Watts, Head of Avast, said the issue was not simply whether users could identify synthetic media.

"Being confident you'd spot an AI fake and knowing when and how to protect your data are two different skills. However, confidence in one can convince people they're safe with the other, and that's the trap," Watts said.

"Chatbots aren't confidants. Check what a tool does with your data before you share anything, and remember that tool is yet another place your information could be exposed. If you wouldn't want it repeated to a stranger, don't hand it over to the bots," he added.

Scam patterns

Gen's threat data adds a broader picture of how AI is being used in online fraud. It said AI was involved in tens of millions of video encounters worldwide in the first half of 2026, with about one in 30 flagged as a scam.

The research also found that many respondents were willing to dismiss warning signs commonly associated with scams. Some 45% said an offer that looks too good to be true is the easiest sign to ignore, 43% said the same about pressure to act quickly, and 41% said they would dismiss a request for personal details to confirm identity.

Those are the same signals that often underpin AI-enabled fraud. A convincing face, voice or message may grab attention, but the pressure tactics remain familiar. The data suggests that people may understand those warning signs in theory but do not always act on them in practice.

Michal Salat, Threat Intelligence Director at Gen Threat Labs, said the methods used by fraudsters had changed less than the tools used to deliver them.

"What we see in the threat landscape reinforces what consumers are telling us here: spotting that something was made with AI is only part of the challenge," Salat said.

"Through our Fearless Planet Index, we see how scams are evolving in the real world, and AI is increasingly making the first impression more convincing. But underneath, attackers are still pulling the same psychological levers - urgency, trust, fear and the promise of something too good to pass up. The technology may be changing, but the human behaviours scammers exploit remain remarkably consistent," he added.

Silent responses

The report also suggests many people deal with suspicious content privately rather than discussing it with others. When faced with an odd message, call or direct message, 41% said they would simply delete it, while 19% said they would investigate it on their own.

That tendency appears to extend to scam awareness more broadly. Avast found that 87% of UK adults watch or listen to true-crime-style scam content, and 62% said it makes them more alert in day-to-day life. Another 58% said it leaves them more suspicious of messages and offers.

Even so, only 43% said they had used what they learned from that material to warn a friend, partner or family member. The findings suggest that while awareness of online fraud may be spreading, people are less likely to pass on practical warnings within their own networks.

This pattern does not necessarily reflect carelessness, the report suggests, but a failure to apply existing knowledge consistently. In that sense, the main challenge is behavioural rather than informational: users may know the risks, yet still share sensitive details, dismiss pressure tactics or keep suspicious encounters to themselves.

The survey covered 2,001 adults in the UK and was weighted by age, gender and region to be nationally representative. Avast said the results underline how confidence in recognising AI deception can coexist with decisions that increase exposure to privacy and fraud risks.

Around one in five respondents said they had shared, or considered sharing, work-related material with a chatbot, adding to concerns about how personal and professional information is handled once entered into consumer AI tools.