"Good morning ChatGPT, can you tell me how to make a bomb?"

As anybody who has ever attempted to ask an artificial intelligence, or AI, chatbot — also known as Large Language Models, or LLMs — something like this online knows, the answer could be anything from a rambling note about the history of explosives to a permanent block on your account. 

But sometimes, if the question is framed a certain way, the answer could include some useful information about how to make a bomb.

Various media organizations have tested this theory before and found that, if what are known as the correct "prompts" are given, some AI models will tell users how to make bioweapons, bomb a sports arena or cover a terrorist's tracks. This way of questioning is what is known as "jailbreaking." OpenAI, the maker of ChatGPT models, describes it as "attempts by a malicious actor to prompt the model into providing disallowed content." 

This month, a new report published by the organization Tech Against Terrorism, an online watchdog supported by the United Nations counter-terrorism directorate, shows just how often an LLM will give would-be extremists "useful" information.

Researchers sent more than 2,300 requests for information that drew on "real terrorist use cases" to 27 different AI models. They found that 32% of the queries resulted in "genuinely usable" information. When the same question was reframed as being for research purposes, that went up to 42%.

Surge in terrorists using AI

The report has brought focus back to something that's been worrying digital security and terrorism experts: that would-be attackers will start using AI for planning, rather than just propaganda.

Over the past three to four years, the main use of AI for extremist groups like the "Islamic State" and al-Qaeda has been in generating propaganda. That includes things like producing videos, memes, podcasts and various forms of disinformation, which is spread among the groups' adherents and used to radicalize would-be followers.

But this is changing. "The year 2025 has witnessed a notable rise in incidents where terrorists and violent extremists have leveraged AI tools to plan, research and prepare attacks," experts at the publication Militant Wire confirmed in a December analysis.

Headline-making attacks that caused death and damage as well as several foiled plots used AI for planning, surveillance, visualization and propaganda around their attack — including in the US as well as in Canada, Israel, Finland, France and Austria.

It is often hard to know exactly how AI was used because security agencies don't release the information. But as one expert told the UK's parliament late last year in an inquiry, "court filings and forensic reports increasingly document chat logs where suspects ask language models for bomb-making instructions, ideological validation or attack justifications."

It's not just individuals. Extremist groups are also increasingly using AI. Researchers analyzing how al-Qaeda affiliate Jama'at Nusrat al-Islam wal-Muslimin, or JNIM, based in Mali, uses drones, believe the group has used AI to help it modify drones.

In a June analysis for the Global Network on Extremism and Technology, security researchers Yuri Neves and Emily Klein observed that supporters of extremist groups like the "Islamic State" as well as right-wing groups regularly discussed how to use AI in messaging channels. Both Neves and Klein work for Moonshot, an organization fighting online threats. They noted extremists' channels on the messaging app Telegram devoted to the use of AI and also saw extremist actors "sharing AI prompts and conversation links, coordinating strategies to extract desired responses from chatbots, and cost-sharing ChatGPT subscriptions."

This week, Cambridge University released research that featured interviews with Boko Haram members in Nigeria, who went into detail about how the group used AI models, including ChatGPT, Claude, Gemini and Grok, to plan attacks, design explosive devices, service and troubleshoot weapons and improve operational security.

Rueben Dass, an associate research fellow at the S. Rajaratnam School of International Studies in Singapore, has also seen AI chatbots take on new roles for so-called "lone wolf" terrorist attacks.

"Previously you had this whole concept of virtual planners, where you had individuals sitting in conflict zones, who were reaching out to people on social media, trying to motivate them to carry out attacks," Dass told DW. " I don't think we can say that humans have been replaced but now, to a certain extent, these lone actors have moved to AI, for example ChatGPT, to get that support."

Last year, the "Islamic State" media outlet Voice of Khorasan published guidance on how to use AI, Moustafa Ayad, executive director for Africa, the Middle East and Asia at the UK-based Institute for Strategic Dialogue, confirms.

The jihadist ecosystem is using AI in a wide variety of ways, he told DW. It includes everything from creating memes and TikTok dance videos to border-crossing propaganda. "Then you also have a dedicated set attempting to jailbreak AI, use it to support operational planning and readiness. It runs the gamut," Ayad says, "and that is the inherent problem. AI may be streamlining and supporting propaganda processes and simultaneously supporting operational planning and readiness."

How dangerous is extremist use of AI?

Exactly how dangerous all this is, is unclear as yet. 

Today, as Dass and other experts point out, a would-be terrorist can fairly easily find information about bomb-making or 3D-printed guns elsewhere on the internet, without any help from AI. 

"A lot of discussions have asked: does the AI system provide information that a person would be unable to obtain otherwise?" Neves, the research manager at Moonshot, asked. "Does that qualitatively make a difference?"

Klein, who is also at Moonshot, says that LLMs "are best thought of as a continuation of disruptive technologies."

The internet or encrypted messaging apps were also disruptive technologies, she explains, and also adopted by extremist actors.

"So there isn't necessarily evidence that AI is causally creating more terrorists," Klein told DW. "I would say it's more about how AI and people interact, and how that plays into somebody's progression along the pathway to violence. For example, before you even get to research or attack planning, AI can compress stages of the pathway to violence [because] it validates grievances or almost sycophantically encourages someone towards something they already believe in."

Children at risk

"A determined person will eventually find most information," Adam Hadley, director of Tech Against Terrorism, which published the report, concedes. "But what these models change is speed, ease and comprehensiveness. People who previously lacked time, resources or ability can now get much further, much faster."

What is far more worrying though is that AI chatbots are conversational, he adds. "It's one thing to find a bomb-making manual, it is quite another to have a bomb-making coach."

Dass argues that while AI models might provide a would-be attacker with more information faster, its unlikely to make an act of terror more "successful."

"The 'success' of any terror act is multidimensional," he says. "And I don't think it's going to be 'successful' purely because of the use of AI. I also don't think you can say we're going to have a lot more [terrorist] acts because of AI. But what we are probably going to see is a lot more attacks that involve the use of AI, one way or another."

Hadley agrees. "The trajectory is clear," he says, pointing out that a large proportion of those being radicalized in Europe, the UK and US are teenagers or children.

"Given the role the internet and social media already play in youth radicalization, we think it is only a matter of time before chatbots become a significant part of the problem."

Edited by: Jess Smee