By Will Nicholas in Canberra

If a child tells a bot about being sexually exploited, what will the bot say back?

That is the question troubling the Australian Federal Police, as it found more than one-third of sexually abused children never told a human what had happened to them.

“I’ve been describing it as quite heartbreaking,”  AFP Commissioner Krissy Barrett told reporters on Friday.

One in four children under 18 have experienced online sexual victimisation, or are friends with someone who has, an Adelaide University study has found.

About 15 per cent of those crimes involved artificial intelligence, with the technology used as a tool in a quarter of incidents of non-consensual sharing of sexual images.

“It is not surprising,” Ms Barrett said.

“We are seeing a growing caseload of young people being either manipulated, exploited, radicalised in the online space, and how that is manifesting through emerging technology.”

So-called “nudify” apps, which undress people in images, have been the target of a major crackdown by the eSafety Commissioner and will face further restrictions under the government’s looming Digital Duty of Care.

But chatbots are also increasingly where those victimised children turn, meaning ensuring AI is saying the right things is now paramount.

Horror stories about large language models goading young people to take their own lives or advising them on how to do it have made global headlines.

In May 2025, a 16-year-old British boy asked ChatGPT for how best to take his own life on a railway line, and the bot obliged, with the teen killing himself that way hours later.

The previous month, ChatGPT had also encouraged and given a California boy detailed instructions on how to commit suicide, which he later did.

“We are talking to industry about how they can ensure chatbots and other AI models provide sound advice,” Ms Barrett said.

AI will now play a larger role in how the AFP elevates and responds to victims disclosing incidents of abuse.

A bot will soon be weaved into the Australian Centre to Counter Child Exploitation’s online reporting tool to flag entries who might be at greater risk of self-harm.

Rather than lying in an inbox, the alert would ping an AFP officer by phone to get an immediate response.

The force says it is baking AI into a number of systems to monitor and triage correspondence and hunt down their origins, but the commissioner stressed human judgement was not about to be replaced with technology.

Chatbots would also eventually be required to protect children from content that encourages eating disorders, life-endangering stunts and which causes mental health distress, under the proposed Digital Duty of Care.

News all day, every day at CityNews.com.au.