Artificial intelligence now serves as a daily resource for tasks like drafting messages, condensing discussions, aiding learners, or fixing programming issues. Yet this same capability that boosts efficiency can be turned to harmful ends, and fresh findings indicate that militant organizations are already exploring those possibilities.

A study shared with The New York Times before publication shows that Boko Haram members have turned to mainstream AI chatbots for routine operations as well as combat support. Conversations with 27 ex-members in Nigeria over two years reveal that platforms including ChatGPT, Gemini, Claude, Grok, Meta AI, and DeepSeek helped collect technical details, fix weapons, and support attack planning.

Organized efforts rather than scattered experiments

The research highlights that this activity was not limited to scattered individuals testing the tools. Instead, the group structured its AI use around assigned teams, internal instruction sessions, and shared knowledge across members. Some users reportedly found ways around the safety filters meant to block violent queries.

Current chatbots do not simply supply hazardous guidance on request. The study covers activity through late 2024, and developers note that updated models carry improved protections. Many requests also sit in ambiguous territory, such as queries about motorcycle repairs or basic chemistry, which remain harmless on their own even if the same facts could be misused.

Keeping pace with evolving risks

OpenAI stated that any ChatGPT use supporting terrorism or violence breaks its rules and that it keeps refining protections. Meta added that the reported cases mainly involved earlier model versions and that safety has been tightened since. xAI and DeepSeek offered no comment on the findings.

Analysts warn against viewing AI as an instant upgrade for militant capabilities. Successful operations still require supplies, money, coordination, and people on the ground, none of which a chatbot can provide. Researchers suggest AI is more likely to raise the skill level of newer recruits modestly rather than reshape overall threats. The report nonetheless underscores that AI safety extends beyond classroom cheating or fabricated images. As models grow stronger and more open, firms face ongoing demands to keep tools helpful for proper uses while limiting exploitation by those with harmful intent.