[PDF] Terrorist Misuse of AI: - The Soufan Center

The Soufan Center has published a report titled 'Terrorist Misuse of AI: A Framework for U.S.-China Cooperation,' the think-tank equivalent of gift-wrapping a grenade. Its central finding arrives with the hedge that keeps lawyers employed: the AQ affiliate JNIM 'appears to have used' open-source AI models to enhance its drone capabilities. Appears. That one verb is doing unpaid overtime — it means somebody found the receipts, just not the sort you'd read aloud under oath, so we shall all proceed as though fog were a finding. Boko Haram, meanwhile, has reportedly had AI guide fighters by furnishing technical know-how, including the design of explosive-release mechanisms. That isn't an intelligence assessment, it's a syllabus.

Note the adjective doing the heavy lifting there: open-source. The free tier of the apocalypse, downloadable wherever there's signal, no purchase order, no licence anyone reads, and a terms-of-service document presumably very firm about drone modifications. Every lab that ships weights and calls it democratisation now has a citation with a 'JNIM' in it, and the acceptable-use policy remains as enforceable as a pub sign saying 'no swearing.' We spent two years being told open weights were a gift to humanity; the gift, it turns out, has been received, and the thank-you note is a Boko Haram bomb mechanism.

Then comes the remedy, and it's a beauty: a framework for U.S.-China cooperation. Two governments that can't agree on the colour of a fire truck are to convene over the one subject the report says they already agree about — that terrorists adopting AI is a significant risk. Congratulations: we've located the last remaining item of Sino-American consensus, and it's the bit where everyone dies. The report cites AI safety frameworks, domestic laws and remarks by PRC and U.S. officials as evidence both capitals take this seriously. Frameworks and remarks. That's the evidentiary standard now: a statute, a framework, and a sentence at a podium. The risk is real, the cooperation is theoretical, and the model that designed the release mechanism hasn't been patched since Tuesday.