Will AI Concerns Become a New Source of Terrorism?

The Issue

Terrorism motivated by opposition to artificial intelligence (AI) remains rare. There have been only two such attacks so far: an April 2026 shooting at the home of an Indianapolis councilor who had backed a local data center project and an attack days later in San Francisco on the home of OpenAI chief executive Sam Altman and the company’s headquarters. Neither attack killed anyone, no organization has been tied to either, and the broader opposition movement to AI remains overwhelmingly peaceful. Other violent activity against AI-related targets has mostly been vandalism and a few serious plots driven by grievances other than AI. This record shows that terrorism motivated by opposition to AI is possible but not that a broader campaign has begun. The threat would grow if attackers began to copy and inspire one another, if extremist communities rewarded violence, or if opponents concluded that peaceful politics cannot stop AI risks.

Introduction

Fears about AI are everywhere. Over 80 percent of Americans think AI development is moving too quickly, and politicians on both the right and left are declaring AI a threat to workers—a fear reinforced by the statements of tech company leaders themselves.¹ Data centers, among the most visible physical artifacts of the AI industry, have drawn criticism for their noise, consumption of water and electricity, and potential effects on local energy prices.² At the same time, existential fears of an AI catastrophe are growing, with computer scientists and industry leaders warning of the dangers posed by advanced models.³

Not surprisingly, these concerns have entered mainstream politics. Congress has held hearings on AI’s effects on the workforce, and many officials and commentators have criticized the Trump administration’s hands-off approach to regulating the industry.⁴ In several 2026 midterm political campaigns, opposition to data centers has become an electoral issue, with candidates attacking opponents over their support for such projects.⁵

New issues enter the political arena constantly, but only a few motivate serious violence. There are, however, early indications that AI-related grievances could be among them.

What follows is a speculative analysis of whether AI-related grievances might become a recurring source of terrorism in the United States. The first section discusses the handful of AI-related incidents that have occurred so far. The second section offers an early assessment of what such incidents reveal about the threat. The third section explains how and why new terrorist causes emerge and what that history suggests for AI-related violence. The report concludes by weighing what the record supports, what it does not, and what would change the threat picture.

Anti-AI Violence So Far

Opposition to AI has rarely crossed into terrorism. This analysis draws on the CSIS U.S. Terrorism Incidents Dataset, in which terrorism is defined as violence carried out by a nonstate actor with the intent to kill and to advance a political objective through the intimidation or coercion of a wider audience.⁶ Of particular relevance to this paper, vandalism and other violence not intended to be lethal fall outside the definition.

So far, only two incidents motivated by opposition to AI qualify as terrorism. There was also a terrorist plot against a data center, but it was not motivated by anti-AI sentiment. This section examines these three incidents, as well as some other anti-AI violent activity that falls below the threshold of terrorism.

Shortly before 1 a.m. on April 6, 2026, someone fired 13 rounds into the front door of Indianapolis City-County Councilor Ron Gibson’s home, where he and his eight-year-old son were asleep. The perpetrator left a note on the doorstep reading “NO DATA CENTERS.”⁷ Five days earlier, the city’s Metropolitan Development Commission had approved a rezoning for a $500 million data center in Gibson’s district over residents’ objections about water, power demands, and pollution—a project Gibson publicly supported.⁸ Police called the shooting targeted, and the FBI joined the investigation.⁹ No arrest has been reported.¹⁰

Four days later, on April 10 at about 3:30 a.m., Daniel Moreno-Gama, 20, of Spring, Texas, threw a Molotov cocktail at a San Francisco residence of OpenAI chief executive Sam Altman, setting the driveway gate on fire. He then went to OpenAI’s headquarters carrying kerosene and incendiary devices, struck the glass doors with a chair, and told security he had come to burn the building down and kill anyone inside.¹¹ Moreno-Gama was carrying a document he wrote titled “Your Last Warning” that called for killing the chief executives and investors of AI companies and listed names and addresses. At one point it read: “Also if I am going to advocate for others to kill and commit crimes, then I must lead by example and show that I am fully sincere in my message.”¹² Moreno-Gama has pleaded not guilty to federal explosives charges and to state charges that include two counts of attempted murder.¹³

A third relevant incident meets the CSIS definition of terrorism but was not motivated by opposition to AI. In April 2021, Seth Aaron Pendley, 28, of Wichita Falls, Texas, was arrested in Fort Worth after taking what he believed were C-4 explosives from an undercover FBI employee. He had spent months planning to bomb Amazon Web Services data centers in Ashburn, Virginia, which he believed held servers for the FBI and the CIA and whose destruction, he believed, could “kill off about 70% of the internet.”¹⁴ He hoped to anger “the oligarchy” into an overreaction that would show Americans they were living under a dictatorship.¹⁵ An FBI agent testified that Pendley was willing to kill any security guard who tried to stop him.¹⁶ Pendley pleaded guilty and was sentenced to ten years in prison.¹⁷

Beyond the three incidents in the CSIS dataset, a broader range of violence against AI-related targets falls short of terrorism but could become a larger problem in its own right. The most serious case was an abandoned plot against xAI’s data center in Memphis. In February 2025, Ethan Paul Early, 25, told a friend that he planned to attack the facility using thermite. Angry at Elon Musk and President Trump, he began buying the ingredients, but friends persuaded him to abandon the plan, and he threw the materials away. One of them then reported him to police.¹⁸ Early pleaded guilty to felony reckless endangerment in July 2025.¹⁹ The plot falls outside CSIS’s definition of terrorism because Early did not demonstrate an intent to kill anyone.

Early’s plot sits at the upper end of the cases outside the dataset. Most of the rest have been simpler vandalism. One of the clearest patterns has involved Flock Safety’s AI-powered license-plate cameras. By early September 2026, vandals had destroyed more than 240 such cameras across the country.²⁰

The fight over AI has also begun to attract existing extremist movements. The Anti-Defamation League recorded members of the white supremacist group European Americans holding a banner reading “A Jewish data center has no home here,” flanked by swastikas, in Howell, Michigan, in January and again in March 2026. A similar banner appeared outside a town hall hosted by a Jewish elected official in nearby Genoa Township in March. That same month, in Houston, the Alamo Youth Club put up stickers reading “No more data centers on our land” alongside antisemitic material.²¹

Early Lessons on AI-Related Terrorism

The incidents described above form a small and uneven record, not a coherent campaign of terrorism. They involve different motives, targets, and levels of violence. Those differences provide the starting point for assessing the threat, but because the number of incidents is small, the following findings should be considered suggestive rather than definitive.

Serious attacks remain rare. Taken together, the record of anti-AI violence is thin: two serious attacks motivated by opposition to AI, one plot against a data center driven by anti-government grievances, and a wider field of vandalism and extremist rhetoric. Opposition to AI and data center development is overwhelmingly peaceful, and although the serious incidents above demonstrate the potential for anti-AI terrorism, three cases do not establish a trend or the scale of the future terrorism threat.

Opposition to AI itself is only one motive for attacking AI-related targets. The incidents so far fall along a spectrum according to how directly the grievance concerns AI technology. Moreno-Gama sits at one end: He attacked because he believed AI would destroy humanity. The Indianapolis shooting emerged from a dispute over the local costs of AI infrastructure. The destruction of Flock cameras reflects opposition to a particular use of AI-enabled technology. At the other end are cases in which the connection to AI is incidental. Pendley saw a data center as a means of crippling the federal government, while Early’s grievance centered on Musk and President Trump rather than AI. The motives of future perpetrators could fall anywhere along this spectrum.

Existing extremist movements could also co-opt anti-AI grievances. The antisemitic imagery in Michigan and Texas show that some white supremacist groups are already doing so, and some anti-government extremists may view data centers as federal infrastructure, as Pendley did. AI also intersects with law enforcement surveillance and environmental concerns, issues that have long motivated extremists. Jihadist movements have not treated opposition to AI as a cause, and no public case so far shows them selecting a target because of its connection to AI. However, jihadists could still come to view AI targets as symbols of Western power.

Anti-AI violence may also draw on a new ideology. The belief that AI threatens human survival already sustains a growing nonviolent movement—one that does not neatly fit current political labels—and one terrorist, Moreno-Gama, already came from its fringe.

The incidents do not yet constitute an organized campaign. No attack or plot described here has been publicly attributed to an organization directing violence against AI-related targets. Moreno-Gama, Pendley, and Early appear to have acted independently, while extremist organizations have limited their activity to propaganda. A shift toward coordination, tactical imitation, or organizational endorsement of anti-AI violence would mark a significant increase in the threat.

People associated with AI may be easier targets than AI facilities. The attacks on Gibson’s and Altman’s homes required only a gun and a Molotov cocktail, respectively, although both were unsuccessful in killing their targets. Data centers and corporate offices may be harder to reach because they have controlled access and dedicated security, and Pendley’s more ambitious bombing plot depended on an undercover FBI employee to supply what Pendley believed were explosives. This disparity may shape future target selection: Attackers unable to penetrate a facility could instead target executives, employees, local officials, or project supporters in locations with fewer security measures. The threat therefore extends beyond the physical security at AI facilities to the people publicly associated with them in their day-to-day lives.

These incidents show what AI-related terrorism has looked like so far but cannot point definitively to a growing trend. Determining that requires examining why some political causes produce sustained terrorism while others remain peaceful.

Why New Terrorist Causes Emerge

Terrorist causes rise and fall, and it is difficult to predict when new issues will produce violence. Nevertheless, several factors may influence whether AI concerns produce more terrorism in the years to come.

The zeitgeist often shapes the emergence of new causes. Cold War rivalries stimulated left-wing terrorism around the world, and Marxist-oriented groups declined at the end of the Cold War.²² White supremacist violence grew after 2008, when the United States elected its first Black president, and during Covid-19 there was a spike in hate attacks against Asian Americans after prominent officials and commentators blamed China for the pandemic.²³ After the 2022 Dobbs v. Jackson Women’s Health Organization Supreme Court decision, which allowed states to ban abortion, the militant group Jane’s Revenge emerged, claiming credit for a string of nonlethal attacks on pro-life pregnancy centers and other targets associated with anti-abortion sentiment. CSIS data indicates that attacks on Immigration and Customs Enforcement (ICE) targets grew in 2025, as President Trump employed ICE in a controversial high-profile campaign to target illegal immigration. In all these cases, a tiny fraction of a much larger political cause embraced violence.

AI itself is a general-purpose technology, and thus it is taking many forms. Individuals may be hostile to certain uses of AI, such as in autonomous weapons systems, external manifestations of compute (like data centers), or effects such as greater unemployment, even while other uses of AI, such as for medical diagnoses, are seen as acceptable. This may lead to numerous smaller constituencies with different AI grievances.

Whether groups and individuals have alternatives to violence, what scholars call the political opportunity structure, also shapes whether they use it.²⁴ Terrorism becomes more likely when supporters of a cause conclude that peaceful politics cannot achieve their goals. For example, in 1960, South African police killed 69 protesters at Sharpeville. The government then arrested thousands of anti-apartheid activists and banned the African National Congress (ANC). Some senior ANC leaders concluded that nonviolent political activity was impossible, and in 1961 they created an armed wing, uMkhonto weSizwe, which conducted sabotage against infrastructure and government targets.²⁵

Group dynamics can also increase the risk of violence. Leaders may initiate violence when they believe a struggle is likely to be difficult and protracted and they fear losing committed members who may become angry or disillusioned.²⁶ Individual terrorists may also take action seeking social approval from a small but radical peer group. Individuals often join such groups to gain a sense of inclusion and belonging, and peer approval—not the broader zeitgeist—is what shapes their behavior.²⁷ One study of jihadist volunteers found that they often joined as groups of friends and relatives who self-radicalized as a group process.²⁸

Counterterrorism can also shape the emergence and strength of different terrorist groups. The failure of U.S. law enforcement to anticipate a surge in white supremacist violence in the 2010s, for example, allowed extremists to spread their views and plot attacks with only limited FBI and police efforts to disrupt them, in contrast to the aggressive post-9/11 focus on jihadist violence.²⁹

Counterterrorism officials may also be slow to recognize the danger. It is always a challenge to allocate resources or otherwise act before a threat emerges. This is particularly likely if the change involves a new type of threat rather than an increased danger from an existing terrorist group or movement.

Each of these pathways produces a different, but at times overlapping, potential path to violence for AI concerns. AI, of course, is part of the zeitgeist, and as such a few zealous anti-AI individuals might embrace violence. They might be more likely to do so if they were part of a broader anti-AI movement, whether in person or online, where they sought to impress the community with their commitment and emerge as heroes. This risk would be even greater if members and leaders of the anti-AI movement believed that progress through electoral politics and protest is not likely to produce success.

Although this brief focuses on how and why AI might become a motivation for terrorism, there are also several countervailing pressures that might limit the threat.

Technology itself is rarely a motivator of terrorism (and, indeed, terrorists are often avid users of the latest technologies, from dynamite to social media).³⁰ Violent social movements have at times emerged after new, disruptive technologies were introduced. For example, during the Industrial Revolution in England in the nineteenth century, the Luddites emerged from mill workers. These workers destroyed textile machines over concerns about lower pay and unemployment.³¹ The anti-nuclear movements that emerged in multiple democracies during the Cold War were largely peaceful, but did at times engage in property destruction and other limited violence.³² Ted Kaczynski (the “Unabomber”) killed three people and injured over twenty in a series of attacks from 1978 to 1995. Kaczynski sent mail bombs to prominent scientists and business executives because of his opposition to industrial and scientific advancement.³³ These movements, however, are rare. Opposition to disruptive technologies has usually been nonviolent, and when it has turned violent, it has more often targeted property than people.

Perhaps more importantly, existing political movements are typically alert to new causes and ways to increase their support, which may give opponents of AI a nonviolent outlet for change. Different political parties and interest groups are well positioned to do this as AI intersects with traditional political debates, such as those over technological innovation and job losses, corporate power, free speech and disinformation, and privacy and surveillance, among other issues. Many Democrats and Republicans have already incorporated concerns over AI into their political platforms, giving opponents of AI more reason to stay within peaceful politics.

Conclusion

Concerns about AI could become a recurring source of terrorism, but the evidence does not show that they have done so yet. The two attacks motivated by AI-related grievances establish that serious violence is possible, but they do not show that a broader campaign has arrived. No organization is directing attacks, nor has there been a wave of lone-actor terrorism. The much larger body of protest and vandalism should not be misconstrued as a precursor to more serious violence.

This assessment would change, however, if a series of isolated attacks began to reinforce one another, such as if perpetrators adopted a shared justification, copied targets or methods, received approval in extremist anti-AI communities, or formed organizations committed to violence. Claims that elections and protest cannot stop AI development would be especially important. A key indicator is not how many people have concerns about AI, but whether any part of that broader opposition begins to regard violence as legitimate and necessary. The challenges for policymakers are therefore to identify any of these transitions early, to protect vulnerable targets, to preserve peaceful avenues for addressing AI-related grievances, and to avoid inflating the threat in response to growing peaceful opposition.

Please consult the PDF for references.

Daniel Byman is the director of the Warfare, Irregular Threats, and Terrorism Program at the Center for Strategic and International Studies in Washington, D.C. Riley McCabe is an associate fellow with the Warfare, Irregular Threats, and Terrorism Program at CSIS.

This report is made possible by general support to CSIS. No direct sponsorship contributed to this report.