CSIS Brief Assesses How AI Shapes Terrorism and Counterterrorism

On 10 July 2026 the Center for Strategic and International Studies published Artificial Intelligence and the Future of Terrorism, arguing that AI is more likely to reinforce existing terrorist trends than to create wholly new forms of attack. The brief also finds that the same tools expand state counterterrorism capacity and calls for independent audits, public-private coordination, CBRN risk revi

Publication

On 10 July 2026 the Center for Strategic and International Studies (CSIS) released a research brief, Artificial Intelligence and the Future of Terrorism. A 29 September 2026 Chinese compilation summarized the brief’s findings on how artificial intelligence intersects with terrorism and counterterrorism.

The authors conclude that AI is more likely to strengthen existing patterns, including decentralization, lone-actor activity, digital propaganda, and information operations, than to produce an immediate new category of terrorism. The same technology is expected to expand state counterterrorism tools, with accompanying privacy and civil-liberties risks. The brief recommends that societies keep technical capacity while putting in place independent audits, public-private coordination, chemical and biological risk review, and legal supervision.

Source: https://www.csis.org/analysis/artificial-intelligence-and-future-terrorism

Cumulative change rather than a break with the past

The brief challenges a common claim that terrorist groups will quickly gain catastrophic new attack capacity, automated lethal systems, or dangerous pathogens solely through AI. It says extreme scenarios should not be ignored, but that focusing only on those scenarios misreads how such groups actually adopt technology.

Historically, terrorist organizations have seldom been first movers in technology. They often have limited resources and fragile structures. Failed experiments can expose networks, cost safe havens, alienate supporters, and draw state retaliation. For that reason many groups prefer methods that are cheap, familiar, and predictable in psychological effect, including firearms, improvised explosive devices, suicide attacks, and vehicle attacks.

Conservatism does not mean stasis. Groups have adopted tools after those tools were commercialized, widely used, or proven by states and firms. Prior examples include social media, encrypted messaging, online propaganda, drones, precision munitions, and information operations. The brief expects AI to follow that path: groups are more likely to use commercial models, open tools, and features already built into ordinary software to cut cost, speed routine tasks, and widen reach, rather than to field their own frontier systems.

Propaganda and recruitment

CSIS identifies propaganda, recruitment, and mobilization as the most immediate uses. Terrorism is treated as violence in the service of attention, fear, followers, and social polarization. Generative systems lower the cost of producing text, images, audio, video, translation, posters, and social posts, so small groups and individuals can sustain high-volume output with fewer people.

The brief also flags audience targeting. Messages can be tailored around religion, anti-government views, ethnic grievance, migration, masculinity, isolation, or conspiracy themes. Language generation and translation reduce the need for dedicated media teams in cross-border campaigns. Synthetic audio and video can impersonate officials, clergy, victims, or fighters in order to fake atrocities, inflate claims, or simulate momentum.

Those effects are described as most acute after an attack, when facts are incomplete and rumors move faster than verification. The authors note that groups may seek confusion, anger, and uncertainty even when audiences do not fully believe fabricated material.

Lone actors and lower-capability groups

The brief gives separate weight to isolated individuals and small cells. Conversational systems can be configured as ideological chatbots that mimic recruiters, mentors, or companions, answering questions and steering users toward more radical material. CSIS states that exposure to extremist content does not by itself produce violence. Radicalization still depends on identity, social ties, politics, psychology, and organizational networks. AI is described as more likely to intensify existing grievances and pathways than to create violent motive on its own.

On operations, the brief says AI can help lower-capability actors organize open-source information, study targets, translate material, draft documents, plan routes, write code, review open-source intelligence, and improve communications hygiene. Much of that information is already public. The change is speed, interpretation, and personalization. CSIS therefore expects a modest rise in the competence of lower-tier actors more than a surge in highly complex plots. Even a small rise is treated as relevant where lone-actor threats already exist.

Financing is expected to move in the same direction. The brief says AI is more likely to improve fraud and deception used for fundraising, including phishing copy, synthetic identities, forged documents, fake charities, and online scams, than to produce novel financial architecture. Groups have long borrowed methods from organized crime and illicit finance. AI is framed as raising the efficiency and credibility of those existing methods.

Extreme-risk claims

The most sensitive section covers chemical, biological, radiological, and nuclear (CBRN) discussion. The authors acknowledge that advanced models may help a malicious actor find technical literature faster or shorten a learning curve in some steps, and that pairing models with drones, image recognition, or open tools could make some preparations more efficient.

They also list hard limits. Biological weaponization is not treated as an information problem alone. It still requires facilities, materials, tacit skill, safety practice, organization, and sustained resources. Model answers are not an operational program. Most groups, the brief says, seek symbolic or political effects and may refuse the uncontrolled harm and retaliation that would follow a mass biological attack.

State sponsors and proxies are discussed in the same cautious register. States with cyber, intelligence, and influence tools may use commercial AI to support proxies with propaganda, open-source analysis, or cyber assistance. Sponsors also face escalation risk and are described as unlikely to hand proxies uncontrolled capacity. The more probable outcome, in this account, is stronger information, cyber, and targeted support, not indiscriminate catastrophic attacks.

A residual concern is a small set of individuals, cult-like groups, and nihilist extremists who seek mass casualties or social collapse. Low probability multiplied by high consequence is cited as a reason for continued monitoring. Policy emphasis is placed on scientific assessment, red-team testing of models, international coordination, and safety audits of systems, rather than on treating every risk as an end-of-world story.

Counterterrorism capacity and five governance steps

CSIS states that AI will also strengthen state counterterrorism. Machine learning can support pattern recognition, anomaly detection, and analysis that inform assessments of time, place, and target type. Agencies may flag unusual procurement, travel, finance, and fragmented leads. Social-media analysis may surface violent advocacy, recruitment, and extremist narratives.

Those uses carry legal and political risk. Terrorism is a low-frequency event, and behavioral prediction is uncertain. A high algorithmic score is not proof that a person will commit violence. Poorly designed systems can amplify bias, raise false positives, expand monitoring, and mislabel people who are not involved.

The brief offers five governance directions:

1. Independent audits of AI counterterrorism systems, rather than reliance on vendor self-assessment. 2. Institutional information sharing and crisis coordination among governments, AI firms, and relevant agencies. 3. Ongoing chemical and biological risk testing, red-team evaluation, and international scientific cooperation. 4. Judicial review, transparent oversight, and accountability so security policy remains inside the rule of law and civil liberties. 5. Rejection of technological determinism, with extremism still analyzed through identity, governance, conflict, organization, and social psychology.

The authors’ closing claim is that AI will not recast terrorism in a single step. It will change the interaction between terrorist groups and counterterrorism states through many small, accumulating shifts. The 29 September 2026 compilation attributed the underlying study to CSIS under the title Artificial Intelligence and the Future of Terrorism.