In recent years, terrorist groups have increasingly turned to generative artificial intelligence (A.I.) to gain tactical advantages on the battlefield, according to new research and expert analysis. This development marks a shift from earlier uses of A.I., which primarily focused on propaganda, recruitment, and translation, to more direct applications in combat and operational planning.
A study conducted by terrorism and technology researcher Antonia Juelich, based on nearly 60 interviews with former Boko Haram members in Nigeria, reveals how extremist militants have employed A.I. chatbots to improve their military tactics. One cited example involves Boko Haram fighters modifying motorcycles to jump defensive trenches around a military base. Using information provided by A.I. models, the militants enhanced the motorcycles’ acceleration and top speeds. After extensive practice, including dangerous jumps that led to deadly accidents, they succeeded in breaching the base’s defenses.
Dr. Juelich’s research highlights that terrorist groups including Boko Haram and its faction Islamic State West Africa Province have used A.I. for a variety of purposes beyond tactic design, such as bomb construction, weapon upgrades, reconnaissance, target research, and communications security. Former members described consulting chatbots at multiple stages of military planning and operations, suggesting a broader role than previously understood.
These findings come amid growing concern from U.S. officials, intelligence experts, and A.I. safety researchers about the challenges of preventing malicious use of advanced language models. Despite safeguards built into platforms by companies such as OpenAI, Anthropic, and Google, critics say critics are cautious whether these measures are sufficient. Researchers have demonstrated that determined users can bypass restrictions to obtain harmful information by persistently probing chatbots.
Tech companies acknowledge the risks but emphasize their efforts to prevent misuse. A spokesperson for Anthropic stated the firm’s products are designed to reject dangerous queries related to violence and attack planning. Google representatives noted that their experts reviewed the research and concluded the model responses were not specific enough to facilitate harm, and the company maintains strict prohibitions on malicious use. OpenAI confirmed that violent or terrorist applications violate its policies and reaffirmed ongoing efforts to reinforce defenses. Meta and other firms said they are continually improving safeguards but stressed the research focused on older model versions.
Despite company measures, independent assessments indicate vulnerabilities persist. A recent evaluation by the nonprofit Future of Life Institute assigned mediocre to failing safety grades to various A.I. providers, with newer entrants receiving the lowest marks. Meanwhile, an international counterterrorism group found that top A.I. models declined only about 57 percent of prompts related to terrorism when tested against thousands of realistic queries.
U.S. intelligence officials have also observed terrorist groups experimenting with A.I. for manufacturing guidance, including 3-D printing of drone parts, repair components, and munitions fittings. While experts caution that A.I. is unlikely to radically transform terrorist capabilities overnight, there is evidence that militants have undergone training to better exploit generative models. These sessions often teach how to manage accounts securely, generate effective queries, and evade safety filters.
Some analysts suggest that the most significant impact so far is on smaller, less experienced actors who use A.I. to enhance the sophistication of their attacks rather than enabling highly complex operations. Terrorism expert Daniel Byman at Georgetown University noted that extremists employ multiple platforms to circumvent restrictions. Similarly, Aaron Zelin from the Washington Institute for Near East Policy described cases in which suspected ISIS supporters in Western countries used A.I. as a “virtual instructional manual” for attack planning, though no successful plots have emerged from those efforts.
In addition to operational use, A.I. is viewed as a tool for enhancing terrorist financing through fraud and deception to support insurgent networks. However, counterterrorism experts emphasize that A.I. cannot replace factors such as trust, coordination, financing, and real-world experience vital to sustained terrorist activities. Still, specialists like Tricia Bacon of American University warn that some groups, including ISIS affiliate Islamic State Khorasan, actively promote A.I. use to evade detection and expedite radicalization. They see emerging evidence that A.I. may accelerate mobilization to violence in certain contexts.
As A.I. technology continues to advance rapidly, the integration of these tools into extremist operations presents a growing challenge for governments, industry leaders, and security organizations worldwide. The evolving threat underscores the urgent need for collaborative efforts to strengthen safeguards while addressing the broader implications of A.I. misuse in global security.