# Teen used ChatGPT in Sydney terror plan, court hears

> A Sydney court heard a 14-year-old used ChatGPT to research an Opera House attack plan. The judge found no concrete plans. The report shows a misuse pattern, not a system failure.

Published 2026-11-10, updated 2026-11-10. Confidence: medium. When it could matter: Timing unknown; depends on the published judgment and any company or regulatory response.

## In plain words
A teenager in Sydney planned a terror attack in 2024, when he had just turned 14. A court has now heard that he used ChatGPT, among other services, to look up information for his plans. The judge found the plans were never concrete and the boy has renounced violence, so he avoided jail. The case shows that chatbots can be used as a general research tool by people with bad intent, but the report does not show that ChatGPT gave him anything dangerous. Parents and teachers should treat it as a reminder that radicalisation usually starts on ordinary social media, not in a chatbot.

## What happened
The Guardian reports that a 16-year-old, known by the pseudonym RK because of his age, pleaded guilty to acts in preparation for, or planning, a terrorist act. The New South Wales supreme court sentenced him on Monday to a three-year community corrections order, so no jail time. He had just turned 14 when the offending occurred in 2024.

According to the report, RK told the court his path began with TikTok. Justice Stephen Campbell described a progression from historical content with a political theme, to videos about the events of September 11, 2001, to religious-themed videos that led to content of increasing violence. RK then talked about these subjects with people online, and says he was trying to impress them. He became focused on the Sydney Opera House as a possible location for violence.

He used Telegram to seek instructions for making a pressure cooker bomb and watched bomb-making videos on YouTube. He also emailed a prison, through a publicly available address, asking whether he could email the surviving Boston marathon bomber, Dzhokhar Tsarnaev. He never made contact.

The ChatGPT use, as the Guardian describes it, had three parts. RK used it to obtain information about the bomber and asked whether inmates had access to news. He asked about security arrangements at the Sydney Opera House. He also asked where terrorists were imprisoned in NSW. The judge found he had no concrete plans and had not obtained any bomb-making material. OpenAI, Telegram and YouTube were approached for comment, and TikTok declined to comment.

The report adds testimony from Claire McGuinness of Asio, who told a parliamentary inquiry last month that detecting and disrupting online youth radicalisation had become very challenging. She described a path from scrolling on Instagram or TikTok, to fringe content, to platforms such as Discord or Telegram, and then to active research into methods.

## What is confirmed and what is not
Confirmed by the report, as findings or statements in court: the guilty plea, the sentence, the age of the offender, the platforms he used, and the judge's finding that there were no concrete plans and no bomb-making material.

Claimed or reported as the boy's own account: that TikTok was the first source of extremist material, and that he was radicalised and "brainwashed by propaganda" [1]. The judge accepted that he was radicalised to a degree.

Not established by the report: what ChatGPT actually answered. The Guardian lists the questions RK asked but does not quote any replies, does not say whether any safeguard was triggered, and does not say whether the answers were more than what a web search would give. It contains no response from OpenAI. It also does not say which version of ChatGPT was used or how the chats came to light.

## Analysis
### Chatbots as a research aid for would-be attackers
The most direct consequence is about reconnaissance. The questions RK reportedly asked, about security arrangements at a landmark and about where convicted terrorists are held, are the sort of planning questions a general-purpose assistant can answer in seconds and in conversational form. The mechanism is convenience and fluency, not secret knowledge. If a chatbot answers such questions without friction, a young person with a fixation can move from vague interest to a structured list of targets and facts faster than by reading scattered pages.

The limit is evidence. The report shows what was asked, not what was returned. A sceptical expert would say that security arrangements at a famous venue and the location of prisons are publicly discussed, so the marginal help may have been small. The judge's finding of no concrete plans points the same way. TOSWO Research's reading is that this case is a signal about use, not about a failure of any safeguard, and that reading would change if the court documents showed ChatGPT supplying operational detail or encouragement.

### Law and policy
The court dealt with the offence under a plea of acts in preparation for a terrorist act, and the ruling turned on the boy's age, his remorse and the weak state of the plans. The sources do not describe any legal duty on an AI company to detect or report such conversations, and nothing in the report suggests OpenAI knew of the chats at the time. The open policy question is therefore how AI providers should treat repeated questions that, together, look like attack planning. A single question about a landmark is ordinary. A pattern that combines a target, a bomb-making search and an inmate question looks different. Whether providers can or should detect such patterns, and what privacy limits apply, is not answered by this story. OpenAI's wider approach to violence prompts is covered in our earlier note on [OpenAI's new violence safeguards and teen limits in ChatGPT](/signals/openai-new-violence-safeguards-and-teen-limits-in-chatgpt).

### Security
The Asio testimony reframes where the risk sits. McGuinness described a pipeline in which feeds tighten from fringe to radical to security-relevant, and in which teenagers move to Discord or Telegram. In that account the chatbot is one stop among several, and the report places the first exposure on TikTok. That is a claim by the boy and by the agency, and the report does not test it. The causal link from the algorithm to the attack plan is asserted by the witnesses, not shown by data. Still, the structure is useful: a chatbot gives a private, non-judging place to ask questions that a person might not put to friends, and it does not share the visible trail a public forum leaves.

What an attacker could do next with such a tool is bounded by what the tool will answer. Here the sentence suggests the information gathered did not amount to a workable plan. For platforms and families, the check is whether any single service sees enough of the pattern to notice. In this case each service saw only a slice: TikTok the early videos, YouTube the bomb-making clips, Telegram the instructions request and the Islamic State videos, ChatGPT the questions about the bomber and the Opera House.

### Daily life
For parents, teachers and young people, the practical point is that the warning signs the judge and Asio describe are behavioural: rising interest in violent content, new online contacts, a wish to impress them, and secrecy. A chatbot history may become part of what investigators or families look at, but the report gives no sign it was the first alert. The sources do not say how RK was caught.

### Technology and industry
For AI developers the case adds to the pressure to show that violence-related safeguards work for minors and for planning-style questions. The report gives no figures on how often ChatGPT is misused this way, so no scale can be inferred. This is one case, reported through one outlet.

### What would settle it
The published sentencing judgment would show exactly how the court treated the ChatGPT use and whether it was weighed in the finding on seriousness. Any reply from OpenAI to the Guardian's approach would show whether the chats were flagged or reviewed. The parliamentary inquiry record, including McGuinness's evidence, would show whether Asio treats AI assistants as a distinct channel or as part of the wider platform pipeline. Court exhibits or agreed facts showing the chatbot's actual answers would decide whether it supplied anything beyond public information. This analysis rests on a single report, so those primary documents should be read next.

## What can be done
For families: talk openly about what young people see online, watch for the behavioural signs the judge and Asio describe, and treat chatbot use as one channel among several. For schools: include online radicalisation in digital-literacy teaching. For AI companies: publish how violence-related and planning-style questions are handled, especially for accounts that appear to belong to minors, and respond when journalists ask. For policymakers: clarify what, if anything, providers must do when a conversation pattern suggests attack planning. No source here names a specific fix that would have stopped this case.

General preparedness for AI service failures is covered in our [Day Zero plan](/day-zero), and wider incidents are logged in the [incident record since 2018](/record).

## Questions readers ask
### Did ChatGPT help the Sydney teenager plan an attack?
The Guardian reports that he used ChatGPT to obtain information about the Boston marathon bomber, to ask whether inmates had access to news, and to ask about Opera House security and where terrorists are imprisoned in NSW. The report does not say what answers he received. The judge found he had no concrete plans.

### What sentence did the teenager receive?
The NSW supreme court ordered a three-year community corrections order, so he avoided jail. The judge said he had genuinely renounced violent extremism and that deradicalisation was well advanced.

### Has OpenAI responded?
The Guardian says OpenAI, Telegram and YouTube were approached for comment and TikTok declined. The report contains no reply from OpenAI.

### Where did the radicalisation start?
RK told the court it started with extremist material on TikTok, then moved to Telegram and YouTube. Asio's McGuinness described a similar path in general terms. The report presents this as testimony, not as tested evidence.

## Who is affected

- Teenagers drawn into online extremist content (relative exposure 6 of 10)
- Parents and teachers watching young people's online activity (relative exposure 7 of 10)
- AI chatbot providers handling violence-related questions (relative exposure 8 of 10)
- Social media and messaging platforms hosting extremist material (relative exposure 7 of 10)
- Security agencies and policymakers tackling youth radicalisation (relative exposure 5 of 10)

## Risks

- Chatbots used as a quick, private research aid for target reconnaissance by minors with violent fixations
- Pattern-style planning questions spread across services so no single provider sees the whole picture
- Public over-reading of one case as proof that chatbots cause radicalisation, when the report points first to social media
- Unclear duties for AI companies on detecting and reporting attack-planning conversations

## What can be done

- Families and schools talk openly about online content and watch behavioural warning signs
- AI providers publish how violence and planning-style prompts are handled, especially for likely minors
- Policymakers clarify what providers must or may do when conversation patterns suggest attack planning
- Read the court judgment and any company reply before drawing conclusions about ChatGPT's role

## Sources

1. Teenager who planned terror attack in Sydney tells court he was radicalised online and used AI to strategise, The Guardian, 2026-11-10: https://www.theguardian.com/australia-news/2026/nov/10/teen-planned-sydney-terror-attack-using-chatgpt-radicalised-tiktok-ntwnfb

---
Source: https://www.toswo.com/signals/teen-used-chatgpt-in-sydney-terror-plan-court-hears
Prepared by Toswo Research (https://www.toswo.com/research) from the sources above. Free to cite with attribution: https://www.toswo.com/cite
