How terrorist groups are using AI to gain an edge in battle

When a gang of Boko Haram members riding motorcycles attacked a military base in eastern Nigeria a few years ago, they were stopped by a defensive moat surrounding the compound.

Extremists have regrouped. They asked the AI ​​for help before launching another attack.

“We saw in the movie how motorcycles can jump over bridges,” the former Boko Haram commander told Antonia Juelich, a terrorism and technology researcher at Cambridge University. “We used AI to learn how to do it. We gave it information like the bikes we use and the distance we need to jump and so on, and that gave us the steps to do what we needed to do.”

Using tips from the chatbots, the mechanics modified the bikes to allow faster acceleration and top speed. The riders dug their own holes, filled them with broken glass and fire, and practiced jumps – sometimes with fatal results – until they achieved enough airborne lift to successfully attack, the defectors said.

The episode, depicted in research paper Dr. Juelich shared with The New York Times ahead of its publication Friday highlights how generative artificial intelligence tools are increasingly helping terrorist groups right on the battlefield, experts say, despite efforts by their creators to protect them from misuse.

Until recently, the Islamic State, al-Qaeda and other extremists used AI mainly in the area of ​​information operations – propaganda production, translation, recruitment and security trading. But that has evolved as jihadists have turned to artificial intelligence to gain tactical advantages in the field, according to current and former US military and counterterrorism officials and independent researchers.

The development highlights a broader challenge for the AI ​​industry. Chatbots have built-in restrictions to prevent users from obtaining information that could harm others or themselves. But researchers have repeatedly found that people can circumvent security protocols, often by slowly but persistently coaxing models to divulge information they’re trained to restrict.

Dr. Juelich has conducted nearly 60 interviews with 27 former members of Boko Haram in Nigeria over the past year. Her field research found that terrorists used chatbots to design explosives, repair or upgrade other weapons, and come up with ideas on how to attack their enemies.

Multilingual models, writes Dr. Juelich, in his report, “were consulted at every stage of military activity—in mission preparation, during operations, and in post-mission analysis—presenting a different picture from the propaganda-oriented use of AI that dominates public discourse and existing public research.”

The research and other recent studies reaching similar conclusions come as concerns grow about the capabilities of advanced artificial intelligence models, which CIA Director John Ratcliffe recently likened to “digital nuclear weapons.” However, the models pose under-recognized risks for other threats, such as creating biological weapons and terrorist activities, AI security researchers and national security officials said.

In recent weeks, the Trump administration has pushed leading labs to let the government review the latest and most powerful platforms before they go public. Government officials largely focus their concerns on the potential for these models to find and exploit software bugs in a way that some fear could wreak havoc on global cybersecurity, not on the potential for terrorism.

“Terrorists are not waiting for us to make AI safe,” said Dr. Juelich in an interview, adding that their use of AI was “significantly underestimated both in scope and character.”

Daniel Byman, Georgetown University terrorism expert and co-author report on AI and the future of terrorism Published on Friday by the Center for Strategic and International Studies, it said terrorist groups are “mixing and matching” different AI systems and trying to avoid technical barriers created by AI companies. Research by Dr. Juelicha also found that Boko Haram is platform agnostic and works interchangeably with OpenAI’s ChatGPT, Anthropic’s Claude, Google’s Gemini and xAI’s Grok, as well as Chinese firm DeepSeek.

The methods described by Dr. The Juelichs typically run until the end of 2024. AI companies have released several iterations of their chatbot models since then, generally saying that while they were more powerful, they also came with stronger security measures. They also noted that some malicious AI features are “dual-use,” meaning that the information shared can also serve legitimate purposes. For example, learning to jump on a motorbike is not inherently harmful or violent.

Other cases described by ex-Boko Haram members, however, seemed more explicitly intended for violence.

“You put in a question or use your voice and you get a detailed answer, like ‘How do I build a bomb?’ and then he tells you how to do it,” said one former commander in the Islamic State West Africa province, Boko Haram’s main faction, Dr. Juelich last year about using an AI chatbot. “It’s like a human robot! We used it a lot.”

Asked about the Boko Haram study, Michael Aciman, a spokesman for Antropic, said the company’s products were “engineered to reject dangerous demands, including those associated with violence, planning attacks, and making explosives.” He added that Anthropic worked with external experts, researchers and industry partners because “no company can face these threats alone.”

Karl Ryan, a Google spokesman, dismissed the research, saying the company’s technical experts reviewed the work and “found that the answers were neither specific nor detailed enough to lead to abuse.” He added that Google has “strict policies against using Gemini to cause real-world harm.” Both Anthropic and Google were informed of the findings by Dr. Juelichem before their publication.

Drew Pusateri, a spokesman for OpenAI, said using the company’s platforms for violence or terrorism violates its policies. “We know that bad actors will never stop trying to abuse our tools, and we will continue to strengthen our defenses in response,” he said.

Meta stated that research by Dr. Juelicha relied on older models rather than the latest version and that he continued to strengthen security measures.

Neither xAI nor DeepSeek responded to requests for comment. Pentagon counterterrorism officials declined to comment on the threat posed by artificial intelligence conspiracies.

Not everyone agrees that safeguards are improving. Non-profit organization Future of Life Institute rated major AI companies on their security commitments this week and concluded that they have mostly eroded across the industry since last year. While most received average marks, xAI and DeepSeek received failing marks.

Other recent studies are consistent with Boko Haram field research. “Artificial intelligence systems can support a range of operational planning functions, including reconnaissance, translation, target research, IED design, itinerary planning, document drafting, coding, communications security, and open source intelligence analysis,” says the Center for Strategic and International Studies report, which refers in part to improvised explosive devices.

Tech Against Terrorism, an international counterterrorism nonprofit supported by the United Nations, released the results last week AI tests measuring how more than two dozen leading models responded to thousands of stimuli derived from real terrorism cases. The tests met with “complete rejection” only 57 percent of the time. While calls for explosives were rejected about 80 percent of the time, improvised chemical weapons were only about a third of the time, the group said.

US intelligence analysts say terrorist groups are also starting to use artificial intelligence to help 3D print weapons parts used in plots, according to a former top US official briefed on the matter. For example, AI is helping some of these insurgents with design and manufacturing instructions for drone components, repair parts and munitions equipment, said the former official, who spoke on condition of anonymity to discuss internal assessments.

Artificial intelligence is unlikely to transform terrorism overnight, analysts and US officials say. Terrorist organizations typically adopt technology cautiously, selectively, and pragmatically.

But the testimony that Dr. Juelich gathered, they depict both eagerness and dedication among Boko Haram cells. Defectors recounted participating in organized training sessions focused on how best to use the capabilities of generative AI models to inform or improve their use of the technology.

The training, in which laptops were equipped with virtual private networks and encryption software, was delivered through transnational jihadist networks, often led by members of the Islamic State, those interviewed said. Common topics included account management on an AI platform, suggestions for generating useful answers, and tips on how to avoid security restrictions.

The examples reveal terrorist networks relying on AI in ways not too dissimilar to how typical office workers incorporate the platforms into their day-to-day work – such as decoding technical information into easy-to-understand steps and displaying online information that would otherwise be difficult to find – albeit with vastly different tasks in mind. Like most of corporate America, terrorist groups seem to have teams dedicated exclusively to working on AI

Some counterterrorism analysts have said that artificial intelligence has so far played a greater role in inspired attack planning by individuals than in larger attacks organized by established groups.

Aaron Zelin, a senior fellow at the Washington Institute for Near East Policy, said his recent research showed that some suspected ISIS supporters in the United States and Western Europe asked ChatGPT questions about potential targets and means of carrying out attacks – a virtual instruction manual. None of the investigations led to a successful plot, he said.

Mr. Zelin pointed to the case of a 27-year-old Tunisian who was arrested in May in relation to a plot that used AI to plan an attack on a museum or Jewish site in Paris.

The Center for Strategic and International Studies report also found that AI is likely to boost terrorist financing primarily by improving groups’ ability to use fraud and deception to raise money to sustain insurgent networks, support individual members, buy equipment and maintain communications.

US officials and researchers have warned that important operational limits remain and that AI will not easily replace the trust, coordination, funding and real-world experience that seasoned terrorist operatives rely on.

“Therefore, the likely result is not a dramatic increase in highly sophisticated attacks, but rather a modest increase in the competence of lower-level actors,” the center’s study concluded.

Still, some analysts have warned of the technology’s reach.

The Islamic State of Khorasan, or ISIS-K, perhaps the group’s most virulent offshoot, is a leader in jihadist circles in urging its followers to use artificial intelligence to avoid detection by authorities, said Tricia Bacon, a Somalia specialist at American University in Washington and a former State Department counterterrorism analyst.

“AI has the potential – and in several cases has demonstrated the ability – to accelerate the process of radicalization and mobilization for violence,” Bacon said.