"'God has helped us, and so will AI': How the Terrorist Group Boko Haram Uses Frontier AI" is a research report by Antonia Juelich, published by the Cambridge Programme on AI Science & Policy (CASP) at the University of Cambridge and circulating from July 13, 2026. It is the first reported study to document institutional-level frontier-AI adoption by a named terrorist organization on the basis of primary interviews with its former members. It anchors AI-Enabled Terrorism.
Scope limitation. CASP publishes an abstract only. The report page carries no body text, PDF, or download link; retrieval itself succeeds (HTTP 200). Everything below is drawn from the abstract, which is reproduced in full in the raw file. Claims about the report's method, evidence base, or reasoning beyond the abstract should not be sourced to this page. Obtaining the full report would require approaching CASP directly.
Method
The study rests on semi-structured interviews with 27 former Boko Haram members, conducted in northeast Nigeria in 2025 and 2026. The abstract characterises the resulting detail as covering AI-assisted activity "primarily through 2024" — a dating qualifier that bounds the findings to a period ending roughly eighteen months before publication. It does not describe sampling, verification, or interview protocol.
Findings reported in the abstract
Both Boko Haram factions use general-purpose frontier chatbots — named as ChatGPT, Claude, Gemini, Grok, Meta AI, and DeepSeek, spanning six developers and both closed- and open-weight models — to assist in "combat and day-to-day operations."
Adoption is described as "institutionalized through specialized units and internal training," rather than ad hoc individual use. Applications named are attack coordination, weapon maintenance guidance, and explosive device engineering. Users are reported to have "successfully circumvented some safeguards."
Transmission of the capability is attributed to "transnational jihadist networks, with Islamic State operatives delivering in-person training" — that is, the know-how moved between organizations through conventional human channels rather than being independently rediscovered.
Respondents are described as expressing "strong enthusiasm for AI and, in some cases, openness to mass-casualty weapons, though documented use remains conventional" — the abstract's own qualifier separating stated intent from evidenced capability. The report's stated conclusion is that "terrorist adoption of AI has thus advanced further and more systematically than prior analysis has recognized," framed as an active threat requiring attention from policymakers, security professionals, and AI companies.
Corroboration
A New York Times investigation published July 10, 2026 independently reported that Boko Haram members use AI chatbots to design explosives, repair and upgrade weapons, and brainstorm attack ideas (Source: nytimes.com).
Relation to frontier-lab threat models
The abstract's finding bears directly on an assumption in frontier-lab safety frameworks — that terrorist organizations remain bottlenecked by technical skill and resources — which underlies the threat models discussed in CBRN Uplift and carried into the GPT-5.6 preview system card (GPT-5.6 (Sol, Terra, Luna)). The documented uses are uplift to conventional operations rather than novel weapons categories, so the finding speaks to the conventional-capability bottleneck rather than to CBRN uplift. That safeguard circumvention is reported across six developers' models bears on whether per-model refusal training removes access at the category level (Jailbreaking and Red Teaming, Open-Weight Frontier Models).
Provenance
Published under the organization's own domain, casp.ac, which serves from a Lovable-hosted application (og:url: aigovernancecam.lovable.app). Verification on 2026-07-14 confirmed title, author, and publisher, and corrected a naming error carried into the wiki from digest coverage: CASP is the Cambridge Programme on AI Science & Policy, not the "Centre for the Study of Armed Groups and Proliferation." Full verification record: Wiki/_meta/queue/gap-scan/proposed-sources/casp-boko-haram-frontier-ai-2026.md.
Relationships
- supports: AI-Enabled Terrorism — the report is that page's primary anchor.
- contradicts: CBRN Uplift — on the assumption that terrorist groups are skill-bottlenecked out of frontier-AI use.
- related: Dual-Use Frontier AI, Jailbreaking and Red Teaming, AI and National Security.