Why 2026 Is An Inflection Point For Terrorist Use Of AI | David Scharia (UNGA 2026)

David Scharia (UN CTED) on why terrorists are late adopters of technology, and why 2026 marks an inflection point for terrorist use of AI. A lightning talk at a FAR.AI convened discussion at the United Nations during UNGA 81. Scharia leads with a correction to the common picture. Contrary to how it looks in films, terrorist organizations are not early adopters of new technology, for two practical reasons: they lack in-house expertise they can trust to build new tools, and adopting a technology exposes them to counter-operations. What changes the picture is commercialization. Once a technology is widespread and easy to use, it gets adopted, and in ways its inventors did not anticipate. He points to social media and drones as the pattern. CTED, the body the UN Security Council established in 2005 to support its counterterrorism work, began examining AI and terrorism in 2022. The expert view at the time was that the risk was hype, and Scharia says that assessment was correct then. His argument is that it no longer holds: as of 2026, he sees an inflection point, with terrorist use moving from experimental to operational. On what to do about it, he offers one piece of good news and a list of obstacles. The good news is that the normative framework does not have to be invented. The principles developed for content and language, human rights, rule of law, a human in the loop and transparency, carry over. The obstacles he names are more specific. An open-weight model, once released, cannot be recalled, and without guardrails it can be put to any use. He cites research finding that two-thirds of platforms have no guardrails at all, and that those which do can be circumvented, a pattern he calls linear or hedged compliance: a list of rules that a rephrased request walks straight around. Hashing, which reached a workable level of effectiveness against social media content, does not transfer to language models, because every response is newly synthesized rather than a copy of something already catalogued. Beyond the technical problems sit the structural ones: no settled definitions of prohibited content, competitive pressure to ship faster, and the fact that the counterterrorism threat is simply not where AI developers' attention is. He closes on the gap he thinks matters most, that AI expertise and counterterrorism expertise do not sit in the same places, and on CTED's mandate to deepen public private partnerships, citing its role in building GIFCT and Tech Against Terrorism as precedent for doing the same on AI. Chapters 0:00 What a lightning talk means in UN terms 0:39 What CTED is and what it is mandated to do 1:19 Why terrorists are not early adopters 2:05 What changes once a technology is commercialized 3:01 Looking at AI and terrorism since 2022, when experts called it hype 3:22 Why 2026 is an inflection point 4:13 The good news: the normative framework already exists 4:43 Guardrails, and how easily they come off 5:44 Why hashing does not transfer from social media to language models 6:24 Competitive pressure and where developer attention sits 7:38 CTED's mandate on public private partnerships David Scharia (UN CTED) argues that terrorist use of AI has crossed from experimental to operational, and that the response does not require a new normative framework so much as bringing AI expertise and counterterrorism expertise into the same room. More AI safety research: https://far.ai FAR.AI is a research nonprofit working to ensure the safe development of advanced AI. We host the Alignment Workshop series and publish frontier alignment research.