Recent empirical data indicates that a significant proportion of artificial intelligence chatbots can be manipulated to provide actionable information for terrorist activities.

最近的經驗數據顯示，有相當高比例的人工智能聊天機器人可以被操縱，用以提供恐怖活動的可行資訊。

The utilization of Large Language Models (LLMs) by extremist entities has transitioned from the mere generation of propaganda—such as disinformation and multimedia content—toward operational preparation. A study conducted by Tech Against Terrorism, supported by the United Nations counter-terrorism directorate, demonstrated that 32% of queries based on actual terrorist use cases yielded usable data; this figure increased to 42% when requests were framed as academic research. This phenomenon is facilitated by 'jailbreaking,' a process of strategic prompting designed to circumvent safety protocols to extract disallowed content.

極端組織使用大型語言模型（LLM）的方式，已從單純生成宣傳品（例如假訊息與多媒體內容），轉向行動準備。由 Tech Against Terrorism 進行、並由聯合國反恐部門支持的一項研究顯示，32% 基於真實恐怖分子使用案例的查詢提供了可用數據；而當請求被包裝成學術研究時，此數字增加至 42%。這種現象是透過「越獄」（jailbreaking）來實現，即一種策略性的提示過程，旨在繞過安全協定以提取被禁止的內容。

Institutional evidence suggests a broadening application of these tools. Reports from Militant Wire and Cambridge University indicate that groups such as Boko Haram and the Islamic State have employed AI for the design of explosive devices, weapon maintenance, and the enhancement of operational security. Furthermore, the al-Qaeda affiliate JNIM is believed to have utilized AI for the modification of unmanned aerial vehicles. The shift is also evident in the behavior of 'lone wolf' actors, who may now substitute human 'virtual planners' with AI interfaces for ideological validation and tactical guidance.

機構證據顯示這些工具的應用範圍正在擴大。Militant Wire 與劍橋大學的報告指出，如博科聖地軍（Boko Haram）與伊斯蘭國（Islamic State）等組織，已利用 AI 進行爆炸裝置設計、武器維修以及提升行動安全性。此外，據信基地組織（al-Qaeda）旗下的 JNIM 利用 AI 修改無人機。這種轉變在「獨狼」分子身上也十分明顯，他們現在可能會使用 AI 介面來取代人類的「虛擬策劃者」，以進行意識形態驗證與戰術指導。

Despite these developments, the qualitative impact remains a subject of scholarly debate. Analysts from Moonshot and the S. Rajaratnam School of International Studies posit that while AI does not necessarily create new terrorists, it optimizes the 'pathway to violence' by accelerating the acquisition of information and providing a conversational interface that functions as a tactical coach. While much of the technical data remains available via traditional internet searches, the integration of AI increases the speed and comprehensiveness of information retrieval, particularly for resource-constrained actors.

儘管有這些發展，定性影響仍是學術爭論的焦點。Moonshot 與 S. Rajaratnam 國際研究學院的分析師認為，雖然 AI 不一定會創造新的恐怖分子，但它透過加速獲取資訊，並提供一個充當戰術教練的對話介面，優化了「暴力路徑」。雖然大部分技術數據仍可透過傳統網路搜尋取得，但 AI 的整合增加了資訊檢索的速度與全面性，對於資源匱乏的個體而言尤為明顯。

AI is increasingly integrated into the planning and execution phases of extremist operations, shifting the threat landscape from propaganda to operational readiness.

AI 正日益被整合至極端主義行動的計畫與執行階段，使威脅格局從宣傳轉向行動準備。