One in every class: AI-assisted sexual abuse widespread among teens

About one in every 25 Australians under the age of 18 has been the victim or has a friend who has been a victim of artificial intelligence-assisted online sexual abuse.

That number — roughly equivalent to one student in every year 12 classroom — comes from the first nationally representative study looking at the role AI is playing in online child sexual victimisation in Australia.

It shows AI has become both a tool for the abuse of young people, but also one of the main avenues to disclose or seek help afterwards.

Adolescents are already turning to AI chatbots such as OpenAI's ChatGPT or Anthropic's Claude to disclose or seek help for online sexual victimisation more often than they do to police or other authorities, the study found.

Tim Cubitt, deputy director of the Australian Cybercrime Observatory at the University of Adelaide, which conducted the study, said the findings showed AI had "well and truly arrived" in the space.

"This is a seismic shift in our understanding of the vulnerability and victimisation of young people online," he said.

First ever nationally representative research on child AI-assisted abuse

The research, a collaboration with the International Centre for Missing and Exploited Children Australia (ICMEC) and the Australian Federal Police's Centre to Counter Child Exploitation, surveyed 1,894 Australians aged 16 to 18.

Researchers asked participants about their experiences of non-consensual sharing of a sexual image and sexual solicitation by an adult.

They found that 15 per cent of young Australians had experienced online child sexual victimisation, including non-consensual sexual image sharing.

About one of in four of those involved AI, the research found.

This typically involved so-called "deepfake" or "nudify" services that use AI to create sexually explicit images of people using real photos as source material. A further one in nine of all teens said that a friend had experienced AI-assisted online sexual victimisation.

Professor Cubitt said these statistics were likely to be a minimum estimate because researchers did not include respondents who said they were not sure whether AI was involved.

He said the study also found when one young person had experienced online sexual victimisation, there was a much higher likelihood someone else in their friendship group also had the same experience.

Boys are major victims of AI-assisted abuse

The study found 19 per cent of girls said they had experienced online sexual victimisation compared with 11 per cent of boys, a finding in line with previous research.

With AI, that pattern flips.

A larger proportion of the online sexual abuse for boys involved AI with 27 per cent of boy's experiences involving the technology compared with just 9 per cent of girls.

Professor Cubitt said that was a significant change from historical patterns of online sexual victimisation.

"This is not something that we've seen before," he said.

Boys were significantly more likely than girls to seek help from AI, regardless of whether they had been victimised, the study found.

More than two-thirds of young people who had experienced online sexual victimisation disclosed it to someone, with the most common recipients being friends (66 per cent) and parents (43 per cent).

The next most common recipient (19 per cent) were AI chatbots, such as ChatGPT or Claude.

That was more common than disclosing to authorities, with police, teachers and other reporting lines (15 per cent).

More than one in five respondents asked for psychosocial counselling from AI each week, and one in 10 disclosed hurtful experiences to AI each week, the research found.

ICMEC Australia chief Colm Gannon called for "an ethical development" for AI so a chatbot did not just accept and forget a disclosure, but then referred people to services

The goal, he said, should be bringing them into a service that could give them a "warm reception".

How AI companies are responding

While all major AI companies have safeguards supposed to stop their tools being used to facilitate abuse, there is little information about their approach to dealing with disclosures or help-seeking around abuse.

OpenAI and Anthropic did not respond to questions by deadline.

OpenAI's public policies prohibit using its services to exploit, endanger or sexualise anyone under 18, including child sexual abuse material, grooming and underaged sexual roleplay.

It says users who attempt to generate or upload child sexual abuse material are banned and reported to the US National Center for Missing and Exploited Children.

Anthropic says it strictly prohibits child sexual abuse material on its services and uses hash-matching on its first-party services to detect known child sexual abuse material in uploaded images, which it reports to the same US clearing house.

OpenAI's ChatGPT has a minimum age of 13, whereas Anthropic's Claude AI prohibits users under the age of 18.

Earlier this year, new rules for AI providers by the eSafety commissioner came into effect that required services to have ways for users to report material — including content depicting self-harm — but they did not set any expectations for mandatory reporting by the services of any disclosure.

Professor Cubitt said the unanswered issue was whether young people who disclosed victimisation to AI systems were receiving appropriate advice and referrals.

More disclosures could improve outcomes for young people, Professor Cubitt said, but only if they were made to a source that could support them.

"There's a real duty of care here among tech companies and AI platforms," he said.

"The crucial next step is to ensure that these disclosures are supported and appropriately referred to the services that these young people want to be engaging with."