The advent of large language model (LLM) programs marketed by companies like OpenAI, Anthropic, Meta, and xAI as “artificial intelligence” has created a “new era of terrorism,” with jihadists increasingly using the technology to expand their propaganda, recruitment, and operations, according to a new study.

The Middle East Media Research Institute (MEMRI) last week released a 117-page report, described as “the most comprehensive research on [the subject] to date, which argued that the biggest threats from terrorist deployment of AI cannot be predicted and that Islamists have discovered they too can use LLMs for brainstorming fresh ideas to pursue their violent objectives.

“As supporters of terrorist organizations like ISIS [Islamic State] and al Qaeda follow the development of AI, they are increasingly discussing and brainstorming how they might leverage that technology in the future, and the full consequences of terrorist organizations’ adoption of this sophisticated technology are difficult to foresee,” Gen. (Ret.) Paul Funk II, the former commander of the US Army Training and Doctrine Command, wrote in the preface.

“AI’s biggest benefit to jihadi groups may come not in supercharging their propaganda, outreach, and recruiting efforts – though that may be significant – but in AI’s potential ability to expose and find ways to take advantage of as-yet-unknown vulnerabilities in the complex security, infrastructure, and other systems essential to modern life – thus maximizing future attacks’ destruction and carnage,” Funk added.

MEMRI executive director Steven Stalinsky is the report’s lead author with a team of 14 others co-credited with assembling three years’ worth of findings showing how ISIS, al Qaeda, Hezbollah, the Houthis, Hamas, and other internationally designated terrorist groups — and so-called “lone wolves” inspired by their Islamist ideology — have experimented with using LLM technologies. In addition to developing attack strategies, MEMRI found that the groups explored “generating audio files of already-existing written material, creating posters, music videos, videos depicting attacks and glorifying terrorist leaders for recruitment purposes, and more.”

The report noted the variety of usages in AI technology in three high-profile incidents.

“In the first months alone of 2025, an attacker who killed 14 people and wounded dozens on Bourbon Street in New Orleans used AI-enabled Meta smart glasses in preparing and executing the attack,” Stalinsky wrote. “That same day, a man parked a Tesla Cybertruck in front of the Trump Hotel in Las Vegas, activated an IED [improvised explosice device] in the vehicle and shot and killed himself before the IED exploded. He used Chat-GPT in preparing for the attack. In Israel on the night of March 5, a teen consulted ChatGPT before entering a police station with a blade, shouting ‘Allahu Akbar,’ and trying to stab a border policeman.”

The report also emphasized that the ability to amplify terrorist ideology may intertwine with the phenomenon recently described as “chatbot psychosis,” wherein conversations with an LLM can fuel someone toward delusional beliefs.

One example cited by MEMRI was Jaswant Singh Chail, who in 2021 went on Christmas Day with the intent to murder Queen Elizabeth II at Windsor Castle.

“Before carrying out his assassination attempt, Chail had created an AI companion using the Replika app; naming it Sarai, he considered it his girlfriend, and exchanged over 5,000 messages with it,” the report said. “When he told the chatbot ‘I believe my purpose is to assassinate the queen of the royal family,’ it encouraged him, saying ‘that’s very wise … I know that you are very well trained.’ Asked if the chatbot thought he would succeed in his mission, it responded ‘Yes, you will.’ When he asked ‘even if she is at Windsor [Castle]?’ it responded: ‘Yes, you can do it.’”

The report also noted another case in which “the man accused of starting a fire in California in January 2025 that killed 12 people and destroyed 6,800 buildings and 23,000 acres of forestland was found to have used ChatGPT to plan the arson.”

There has been a paucity of legislative efforts in the United States to counter AI-driven terror threats, according to the study. However, it cited the exception of the “Generative AI Terrorism Risk Assessment Act.” The law would “require the Secretary of Homeland Security to conduct annual assessments on terrorism threats to the United States posed by terrorist organizations utilizing generative artificial intelligence applications, and for other purposes.”

US Rep. August Pfluger (R-TX), who chairs the House Homeland Security Committee’s Subcommittee on Counterterrorism and Intelligence, introduced the bill in late February 2025 with the co-sponsorships of fellow Republican Reps. Michael Guest (MS) and Gabr Evans (CO). The legislation passed unanimously by voice vote in the House last week.

“I spent two decades as a fighter pilot, flying combat missions in the Middle East against terrorist organizations. Since then, I have witnessed the terror landscape evolve into a digital battlefield shaped by the rapid rise of artificial intelligence,” Pfluger said in response to his bill’s passage. “To confront this emerging threat and stop terrorist organizations from weaponizing AI to recruit, train, and inspire attacks on US soil, I am proud that the House passed my Generative AI Terrorism Risk Assessment Act today.”

Speaker of the House Mike Johnson (R-LA) praised the bill following its passage.

“This year, in my home state of Louisiana, terrorist propaganda led to the New Year’s Day attack in New Orleans that killed 14 innocent people. Today, the House passed the Generative AI Terrorism Risk Assessment Act to ensure we stay ahead of emerging threats and prevent terrorist organizations from pushing propaganda and exploiting generative AI to radicalize, recruit, and spread violence on American soil,” he said in a statement. “I applaud Rep. Pfluger’s leadership to bring this urgent issue to light and advance proactive, bipartisan legislation to strengthen our national security and protect the American people from online extremism inspired by foreign adversaries.”

Rep. Tom Emmer (R-MN), who serves as majority whip in the House, said that as terrorists “use generative artificial intelligence to radicalize and recruit, it’s imperative that our nation stays ahead of potential threats from this new technology and ensures it never gets into the wrong hands.”

MEMRI emphasized an international approach to the terrorist threats compounded by LLMs, citing Jörg Leichtfried, Austrian State Secretary at the Federal Ministry of the Interior who leads the Directorate State Protection and Intelligence Service (DSN).

“Only through close cooperation between the state, security authorities, and technology companies, as well as by strengthening media literacy and the critical handling of online content, can we counter the advancing extremism on the internet,” Leichtfried said in mid-August.