Terror 2.0

This flagship session explored how generative artificial intelligence is transforming recruitment, propaganda, and operational planning, and considered implications for U.S. AI governance and national security frameworks. It featured Michael Mieses, Cyber Intelligence Consultant and Nicholas Bellante, Principal Consultant at CrowdStrike; moderated by Dr. Omar Mohammed, Senior Research Fellow at the Program on Extremism.

On June 11, 2026, the Program on Extremism at The George Washington University hosted “Terror 2.0: Redefining Terrorism and Extremism in the Age of AI,” as part of its Global Extremism Papers webinar series. The session was moderated by Dr. Omar Mohammed, Senior Research Fellow at PoE, and featured Michael Mieses, a cyber intelligence consultant with the Army National Guard background, and Nicholas Bellante, Principal Consultant at CrowdStrike. Mieses and Bellante discussed a paper that they co-authored with Dr. Nakissa Jahanbani, which examines how artificial intelligence is poised to redefine terrorism and counterterrorism in the coming decades.

Mieses and Bellante argued that AI functions less as a wholly new weapon than as an accelerant, fusing cyber, kinetic, and psychological capabilities and lowering the barrier to entry for threat actors ranging from lone wolves to transnational crime organizations. Drawing on historical hybridization examples - Colombia's FARC and ISIS in Syria and Iraq - the authors showed how non-state actors have long blended illicit revenue streams and propaganda to further their agendas, and argue AI represents the newest, most accessible layer in that same pattern.

The paper identified five capability domains where AI is accelerating existing terrorist tactics rather than creating entirely new ones: AI-enabled recruitment, identity deception, physical and kinetic weapons (particularly repurposed commercial drones), AI-enhanced cyber operations, and chemical/nuclear tactical innovation. Analysis of open-source data showed a near doubling of reported AI-related incidents between 2023 and 2025, with a peak in synthetic media around the 2024 U.S. election cycle and a 2025 surge in jailbroken models used for cyberattacks. The authors presented case studies including the 2022 Zelenskyy deepfake, a 2025 incident in which an AI coding agent mapped attack chains against 17 organizations, disinformation during the 2025 Philippine election, and AI-guided drone warfare in the Russia-Ukraine conflict.

The authors projected a three-phase threat horizon - normalization (5–10 years), escalation (11–15 years), and catastrophic plausibility (15+ year) - cautioning that the last phase describes what becomes technically possible absent guardrails, not a prediction.

Mieses noted that AI differs from earlier technological adoptions by terrorist groups because it is the first technology to simultaneously accelerate cost, time, and reach across information, cyber, and kinetic domains at once, while also complicating legal attribution frameworks built around identifiable actors with declared intent. Bellante added that AI's core effect is lowering the skill barrier for who can carry out an attack, not just extending existing capabilities.

Dr. Mohammed raised a recent New York Times report on researchers successfully prompting an AI chatbot into producing bioweapon synthesis instructions, asking how such vulnerabilities could be addressed. Both authors pointed to regulatory sandboxing, citing Anthropic's approach with its Mythos model, as a promising, if imperfect, model for pre-release red-teaming. On the feasibility of public-private cooperation and AI regulation more broadly, both Mieses and Bellante acknowledged the difficulty but argued that fusion-center-style, cross-sector partnerships offer the most realistic path forward, distinguishing between regulating “onshore” Western models and controlling access to compute and infrastructure for offshore, open-weight models.