A recent report by Tech Against Terrorism highlights concerns that AI systems, such as large language models (LLMs), can provide actionable guidance to individuals seeking to carry out acts of violence. The study involved testing over 2,300 prompts related to 'real terrorist use cases' across 27 AI models. It found that 32% of these prompts yielded 'genuinely usable' information, increasing to 42% when the questions were rephrased as research inquiries. Researchers noted that while AI has previously been used by groups like Islamic State and al-Qaeda primarily for propaganda, there is now evidence of its use in planning and preparing attacks. Examples cited include the use of AI in planning attacks in the U.S., Canada, Israel, Finland, and Austria, though specific details remain unclear due to limited disclosure by security agencies.
Bias read (Center): The article presents findings from multiple independent analyses and reports without overtly favoring any side. It discusses potential risks associated with AI usage by extremist groups but does not take a stance on whether these risks are overstated or understated. The framing remains neutral, with
