‘Just ask Grok’: How ISIL is using Big Tech’s AI to build bombs

Cambridge researchers found that Boko Haram factions, including ISIL affiliate ISWAP, used major AI chatbots including ChatGPT, Claude, Gemini, Grok, Meta AI and DeepSeek for bomb-making, combat planning, drone modification and post-attack tactical analysis during 2023-24. Roughly 10 of 27 former fighters interviewed described AI use involving explosives, while ISIL supporters were also sharing methods to jailbreak model safety controls. The findings heighten regulatory and reputational risks for AI platforms as researchers warn that model capabilities are advancing faster than safeguards, amid a US policy stance favoring minimally burdensome AI regulation.

Analysis

The investable issue is not near-term revenue loss; it is a higher probability of forced safety spend, product friction and liability scrutiny just as frontier-model competition is pushing companies to relax guardrails and shorten release cycles. GOOG has the larger enterprise and public-sector exposure, where procurement teams can convert safety concerns into delayed deployments and more onerous indemnification requirements. META is more exposed to reputational and policy risk because broad model distribution reduces control over downstream use, potentially raising the discount rate applied to its AI-driven engagement and advertising upside.

Over the next 1-3 months, the likely catalyst chain is congressional or European regulatory inquiry, followed by demands for independently audited misuse testing, incident reporting and user-verification controls. These measures are manageable for hyperscalers financially but can widen the moat versus smaller model developers: compliance, red-teaming and provenance infrastructure become fixed costs, favoring GOOG, META, MSFT and AMZN relative to private or China-linked challengers. The more material 6-18 month second-order beneficiary is cybersecurity: enterprises and governments will spend on identity, content provenance, model monitoring and threat intelligence as AI lowers the operational skill threshold for adversaries.

Consensus may overreact to headline liability for large platforms while underpricing the regulatory advantage of scale. A sustained de-rating requires evidence that controls impair user growth, cloud AI bookings, or model-release velocity—not merely policy rhetoric. The key falsifiers are no formal enforcement action, no disclosed material increase in trust-and-safety expense, and continued AI-product monetization or cloud backlog acceleration at upcoming earnings.

Key Decisions for Investors

• Avoid directional shorts in GOOG or META solely on this development; use any 3-5% policy-driven weakness over the next month to accumulate GOOG versus a basket of smaller AI software names, as compliance costs should consolidate enterprise AI demand toward scaled providers. Exit the relative-long thesis if Google Cloud AI backlog or enterprise-contract conversion decelerates materially at the next two earnings reports.
• Initiate a 6-12 month long PANW / short IGV pair in modest size. The thesis is that AI-enabled threat activity shifts security budgets toward platform consolidation and automated detection while generic software multiples remain vulnerable to higher compliance and AI-disruption risk; target 10-15% relative return, with a 7% stop on relative underperformance.
• Watch META for a regulatory-premium entry rather than chase a selloff: purchase 6-month downside protection only if implied volatility remains below its 12-month median and the stock fails to recover after a formal US or EU inquiry. The hedge is justified if enforcement introduces distribution restrictions or verifiable trust-and-safety cost escalation; absent those triggers, a standalone put is low-conviction.
• Monitor procurement language from US federal agencies, the EU AI Act implementation calendar, and disclosed safety-capex guidance from GOOG/META. A mandate for independent model evaluations or provenance standards would be a structural positive for incumbents and security vendors, but a broad ban on open-weight distribution would make META the more vulnerable leg.