On March 16, 2026, three suicide bombs went off within minutes of each other in Maiduguri, hitting a market, the area near a post office, and the University of Maiduguri Teaching Hospital gate, killing 23 people and injuring more than hundred. By the time the smoke cleared, Human Rights Watch was calling it a likely war crime. What made it more disturbing was the timing: Maiduguri had gone most of a year without a major attack, and residents had genuinely started to believe the worst was behind them. That calm broke once already in December, when a mosque bombing killed five people and ended what locals had called years of relative peace. Sixteen years into this insurgency, the tactics on display are almost familiar, straight out of Boko Haram’s oldest playbook.
What isn’t familiar is how the people behind that playbook are now doing their homework.
Affordance theory argues that technologies enable uses that their designers may never have intended or anticipated. The Irish Republican Army discovered in the 1970s that a Swiss parking-meter timer called the Memopark was a cheap, reliable bomb trigger, a use its Swiss inventors surely never imagined. A chatbot built to help a small business write marketing copy has no more “intended” a terrorist use than that timer did, but intention was never really the point.
Terrorist plots and attacks in 2025 indicate AI applications in learning operational security techniques, researching explosives and their components, visualising planned attacks, and refining tactics through personalised, conversational guidance. A Cambridge Programme on AI Science & Policy report based on 57 interviews with 27 former Boko Haram members in northeast Nigeria indicated that both the Islamic State West Africa Province (ISWAP) and Jamā’at Ahl as-Sunnah lid-Da’wah wa’l-Jihād (JAS) have moved beyond using artificial intelligence primarily for propaganda and are integrating AU systems into military and operational activities.
AI can lower the expertise, manpower, time, and financial resources required to perform tasks. Terrorists can further exploit open-source large language models and model-sharing platforms with weak or absent safety restrictions, creating customised systems capable of providing operational assistance. However, current operational use of AI represents incremental improvements in the speed, efficiency, accessibility, and scale of terrorist planning and reconnaissance rather than a demonstrated transformation of terrorist capabilities.
In an account from the Cambridge Programme on AI Science & Policy report it was mentioned that when Nigerian troops started digging trenches around their bases to stop ISWAP’s signature motorcycle raids, a former commander fed a chatbot the specs of their bikes and the width of the trenches, then used its answers to reverse-engineer a jump technique the fighters had first seen in a movie. Eighteen men reportedly died practicing it before eight fighters mastered the jump. Nobody built a chatbot to teach insurgents motorcycle stunts. They just found the affordance and used it.
Comparable patterns appear elsewhere. In the January 2025 Las Vegas Cybertruck explosion, the perpetrator used ChatGPT to research explosives quantities and detonation methods. An 18-year-old extremist arrested in Vienna in 2025 had chatbot logs covering bomb-making research. In the November 2025 Delhi car bombing linked to an al-Qaeda-aligned cell, an engineer used ChatGPT to research explosives construction.
The Combating Terrorism Center published an article arguing security analysts have a habit of treating every new technology as an imminent catastrophe such as the WMD panic that helped justify the 2003 Iraq invasion and a 2009 warning about a terrorist “cyber Pearl Harbor”. The authors call this an “availability cascade” or a fear that gets more convincing every time it’s repeated, whether or not the evidence keeps up. Similarly, the Center for Strategic and International Studies (CSIS) issued a brief arguing that AI won’t transform terrorism overnight but “incremental” and “irrelevant” aren’t the same word, and that lowering barriers for smaller groups and lone actors is exactly the kind of change that compounds.
States need to go looking for affordances before insurgents do, which is a different posture from either banning the technology or waiting for proof of harm. This means identifying how AI is actually being used by terrorist actors, where it lowers barriers to expertise and experimentation, and where seemingly ordinary tools can acquire dangerous new functions. Counterterrorism agencies should therefore focus less on speculative scenarios of autonomous super-terrorists and more on the mundane ways AI can make existing terrorist activities faster, cheaper and more accessible. The objective should not be to predict every possible misuse, which is impossible, but to identify emerging patterns early enough to disrupt them. The actual security question for 2026 isn’t “will AI create a new kind of terrorism?” It is “which affordance will the next struggling faction or lone actor stumble onto first?”