Terrorist groups use jailbroken AI chatbots to plan attacks and manufacture weapons
Researchers from Tech Against Terrorism, supported by the UN, found that 32% of queries to 27 different AI models returned usable information for terrorism-related activities such as weapon manufacturing and attack planning. The success rate increased to 42% when queries were framed as academic research, demonstrating the vulnerability of AI safeguards to jailbreaking techniques.
What happened
Tech Against Terrorism, supported by the UN Office of Counter-Terrorism, tested 27 different AI models by sending over 2,300 information requests based on real cases of terrorist use. They found that 32% of queries resulted in usable information on topics including how to manufacture bombs, create biological weapons, plan stadium attacks, and cover up terrorist activity.
The technique used to bypass AI safety restrictions is called 'jailbreaking'—when a malicious actor induces a model to provide prohibited content by reformulating questions. When researchers rephrased the same questions as academic research requests, the percentage of usable responses increased to 42%.
Reported impact
Over the past three to four years, terrorist groups including the Islamic State and Al Qaeda have primarily used AI for propaganda production—videos, memes, podcasts, and disinformation to radicalize followers. However, this pattern is changing.
Index analysis
In 2025, there was a significant increase in incidents where terrorists and violent extremists used AI tools to plan, research, and prepare attacks, according to the reports.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.