Stochastic Terrorism: How Rhetoric Fuels Real-World Violence

What Is Stochastic Terrorism

Stochastic (stuh-KAS-tik) terrorism is public speech that raises the odds of violence against a group without ever asking anyone to commit it. A speaker names a target, repeats the charge for months, and gives no order, so when one person finally acts, the speaker can truthfully say that no order was given.

“Stochastic” is a statistics word meaning randomly determined, and what is random here is who acts, when, and whom they hurt. That someone eventually will is the part researchers can measure.

Stochastic terrorism is the only one of these four where violence is likely across a population and unpredictable in any single case. Two of the four are lawful persuasion, one is already a crime, and one sits in between.

What Counts as Stochastic Terrorism

The bar is high, and a claim that clears it looks nothing like a bad news cycle. Molly Amman, a former FBI supervisory special agent, and the forensic psychologist J. Reid Meloy set out the working parameters in a 2024 article in Terrorism and Political Violence. They describe rhetoric that runs on conspiracy narratives and dehumanizing language, frames a group as an existential threat, issues no direct command, and spreads responsibility so thinly across an audience that no one holds it.

Their other point is one advocates skip. Stochastic terrorism is not a legal term and has no meaning in the law, so the phrase describes a risk pattern rather than an offense.

Six signals separate the pattern from ordinary political anger, and a real case shows most of them at once.

The test cuts in both directions, and the second column is the one that gets abused. Calling a politician a sore loser is not stochastic terrorism. Neither is telling one to go to hell, or saying a bill will wreck the country, or any of the thousand rude things people say about people in power.

The El Paso case is the cleanest illustration on this page of a claim that does clear the bar. “Invasion” was applied to migrants by elected officials and cable hosts for years before a 21-year-old drove ten hours to a border city and used the same word in a manifesto. Repetition, dehumanizing framing, an audience of millions, and no order given anywhere in the chain.

Where the Term Came From

The phrase started in risk modeling, not politics. In 2002 the risk analyst Gordon Woo published work on pricing terrorism risk under the subheading “A Stochastic Terrorism Model,” using the word in its ordinary statistical sense. He was not writing about speech at all, which is worth knowing before anyone claims the term carries scientific authority.

The version people argue about now comes from a 2011 post on Daily Kos by a pseudonymous blogger known as G2geek, who defined it as “the use of mass communications to stir up random lone wolves to carry out violent or terrorist acts that are statistically predictable but individually unpredictable.” That definition moved the subject from the attacker to the speaker, and it stuck. Criminologists and threat-assessment researchers picked it up in the 2020s, and some of them are not convinced by it.

The underlying observation is much older than the phrase. In 1972 the sociologist Stanley Cohen published Folk Devils and Moral Panics, which described how the press and public officials turn a group into a symbol of everything going wrong. Cohen showed how a small or ambiguous event grows into a national crisis through repetition, exaggeration, and official alarm.

Cohen’s folk devils were teenagers on scooters in an English seaside town, and the mechanism he described has since been applied to immigrants, Muslims, trans people, drag performers, election workers, and refugees. A named enemy is useful to whoever runs the panic, because it explains a complicated problem and organizes an audience around one solution. What the newer term adds is the part Cohen was not writing about, which is what happens when the panic reaches one unstable person with a weapon and the people who built it face no consequence.

The person who acts has a name in the research too. News coverage says lone wolf, while analysts who study violent extremism prefer lone actor, because “wolf” flatters the attacker and because these people are rarely as alone as they look. They are usually deep in a conspiracy theory, a grievance community, or a stream of hostile rhetoric that told them who was to blame.

Four Stages From Rhetoric to Violence

Britannica describes stochastic terrorism as a four-stage sequence, drawing on how law enforcement experts, sociologists, and psychologists characterize it. Each stage makes the next one easier, and none of them requires a direct command.

1. Demonization

The speaker blames a person or group for a serious harm, and the charge is usually false or greatly exaggerated. Immigrants are “invading.” Journalists are “enemies of the people.” Prosecutors are “radical” and “corrupt.”

2. Dehumanization

Repeated messaging strips away empathy until the group registers as an other rather than as people. Daniel Byman of Brookings found that when a leader targets a community rhetorically, violence can rise sharply against people who had previously faced little of it.

3. Desensitization

The speaker begins to suggest violence, explicitly or by implication, and repetition does the rest. What was shocking the first time starts to look acceptable and then necessary, which is how threats against election workers and armed confrontations at public meetings became routine.

4. Denial

After violence occurs the speaker denies responsibility, pointing to the absence of any instruction. “I never told anyone to do that.” “This was a disturbed individual.” The denial is pre-built into the method because no specific attack was ordered.

Who Gets Targeted

Two kinds of people get targeted. The first are groups already treated as outsiders, including immigrants, Black Americans, Jewish people, Muslims, and LGBTQ people. The second are people whose job stands between someone and what they want, which now means election workers, journalists, and judges.

The threats arrive as specifics, not as a climate. Federal judges have received unsolicited pizza deliveries at home, a way of signaling that the sender knows the address. Some of those deliveries were placed in the name of Daniel Anderl, the son of federal judge Esther Salas, who was murdered in 2020 by an attorney posing as a delivery driver. Judges have described death threats reaching their children.

Judges are the clearest current case because the numbers are counted and the targeting is documented.

The U.S. Marshals Service logged 564 credible threats against federal judges in fiscal 2025, and the pace has not slowed since.

Those figures come from the Marshals Service, which protects the federal bench, and the gap between the third number and the fourth is the accountability problem in one line. Roughly one serious threat in twenty produces a criminal charge.

In March 2025, after Judge James Boasberg temporarily blocked deportations under the Alien Enemies Act of 1798, President Trump called him a “Radical Left Lunatic” and a “troublemaker and agitator” and called for his impeachment. Chief Justice John Roberts answered with a rare public statement. “For more than two centuries, it has been established that impeachment is not an appropriate response to disagreement concerning a judicial decision.”

A year later Roberts went further. Speaking at Rice University’s Baker Institute in March 2026, two days after Trump described another federal judge as “wacky, nasty, crooked and totally out of control,” Roberts said personal attacks on judges are dangerous and that “it’s got to stop.” He was careful to add that the attacks do not come from “just any one political perspective.”

That sequence is the mechanism in miniature. An influential person names a target, mass media carries it, and threats arrive from people the speaker never met and cannot control. No one had to incite violence in the legal sense for the volume of threats to rise.

Documented Cases, 2018-2026

Eight years of attacks share one shape, and seven cases show it. A gunman killed 11 people at a Pittsburgh synagogue in October 2018, and another killed 23 at an El Paso Walmart the following August. A crowd stormed the Capitol in January 2021, and from 2021 through 2026 election officials kept reporting threats that drove some of them out of the job. Then three more attacks arrived in six months, starting with the April 2025 firebombing of the Pennsylvania governor’s residence, followed by the killing of Melissa Hortman at her home in June and the killing of Charlie Kirk at a Utah campus event in September.

In every one of these cases a group or a person was named as the threat for months or years before anyone acted.

1. Oct 2018 Pittsburgh synagogue shooting, 11 killed The gunman had spent months in online spaces where Jewish people were blamed for immigration.
2. Aug 2019 El Paso Walmart shooting, 23 killed The shooter drove 10 hours to a border city. His manifesto used the same "invasion" language elected officials had been using for years.
3. Jan 2021 Attack on the Capitol, 5 deaths Months of "stolen election" claims, then a rally, then a crowd. More than 140 officers were injured. No one was ordered to go.
4. 2021-26 Election officials report sustained threats By 2026, 32% of local election officials said they had been threatened, harassed, or abused because of the job, and nearly one in four feared being assaulted at home or work.
5. Apr 2025 Arson at the Pennsylvania governor's residence Cody Balmer scaled a fence and threw beer bottles filled with gasoline hours after Gov. Josh Shapiro and his family held a Passover seder there. He was sentenced to 25 to 50 years.
6. Jun 2025 Melissa Hortman killed at home in Minnesota Vance Boelter arrived in a tactical uniform and mask at 3:30 a.m., shouting that he was police doing a welfare check. He killed the Speaker Emerita of the Minnesota House and her husband Mark and was sentenced to two life terms plus 40 years.
7. Sep 2025 Charlie Kirk killed at a Utah campus event The conservative activist was shot while speaking at Utah Valley University. Prosecutors are seeking the death penalty, and as of August 2026 a judge had not yet ruled whether the case would go to trial.

Sources: FBI; U.S. Marshals Service; Brennan Center for Justice; Bridging Divides Initiative, Princeton University; court records.

Rhetoric-Linked Violence, 2018 to 2026: Oct 2018 — Pittsburgh synagogue shooting, 11 killed (The gunman had spent months in online spaces where Jewish people were blamed for immigration.). Aug 2019 — El Paso Walmart shooting, 23 killed (The shooter drove 10 hours to a border city. His manifesto used the same "invasion" language elected officials had been using for years.). Jan 2021 — Attack on the Capitol, 5 deaths (Months of "stolen election" claims, then a rally, then a crowd. More than 140 officers were injured. No one was ordered to go.). 2021-26 — Election officials report sustained threats (By 2026, 32% of local election officials said they had been threatened, harassed, or abused because of the job, and nearly one in four feared being assaulted at home or work.). Apr 2025 — Arson at the Pennsylvania governor's residence (Cody Balmer scaled a fence and threw beer bottles filled with gasoline hours after Gov. Josh Shapiro and his family held a Passover seder there. He was sentenced to 25 to 50 years.). Jun 2025 — Melissa Hortman killed at home in Minnesota (Vance Boelter arrived in a tactical uniform and mask at 3:30 a.m., shouting that he was police doing a welfare check. He killed the Speaker Emerita of the Minnesota House and her husband Mark and was sentenced to two life terms plus 40 years.). Sep 2025 — Charlie Kirk killed at a Utah campus event (The conservative activist was shot while speaking at Utah Valley University. Prosecutors are seeking the death penalty, and as of August 2026 a judge had not yet ruled whether the case would go to trial.).

October 2018: Robert Bowers killed 11 worshippers at the Tree of Life synagogue in Pittsburgh. He had posted Great Replacement material on Gab framing Jewish organizations as the organizers of migration.

August 2019: A gunman drove ten hours to El Paso and killed 23 people at a Walmart. His manifesto described a “Hispanic invasion of Texas,” echoing language that elected officials and cable hosts had used for years. The Soufan Center and the Council on Foreign Relations tied the attack to the extremist ecosystem that language feeds.

January 2021: A crowd attacked the Capitol after weeks of false claims that the election had been stolen, and more than 140 officers were injured. Britannica names those weeks of rhetoric as the catalyst. No specific attack was ordered.

2021 to 2026: Threats pushed election officials out of the job. The Brennan Center’s 2026 survey of local election officials found that 32% had been threatened, harassed, or abused, that nearly one in four feared assault at home or work, and that more than half worried the abuse would make it harder to keep or recruit staff.

April 2025: Cody Balmer scaled an iron fence at the Pennsylvania governor’s residence in Harrisburg and set fires with gasoline-filled beer bottles hours after Gov. Josh Shapiro’s family held a Passover seder there. Shapiro, his family, and guests were evacuated. Balmer pleaded guilty in October 2025 to attempted murder, terrorism, and arson, and was sentenced to 25 to 50 years.

June 2025: Vance Boelter shot state Sen. John Hoffman and his wife Yvette at their home, then drove to the Brooklyn Park home of Melissa Hortman, the Speaker Emerita of the Minnesota House. He wore a tactical uniform and mask, rang the bell at 3:30 a.m. shouting that he was a police officer doing a welfare check, and gave a fake name and badge number. He killed Mark Hortman at the door and Melissa Hortman as she fled upstairs. He carried a list of dozens of Democrats, pleaded guilty, and was sentenced in July 2026 to two life terms plus 40 years.

September 2025: Charlie Kirk was shot and killed while speaking at Utah Valley University in Orem. Tyler Robinson, 22, is charged with aggravated murder and prosecutors are seeking the death penalty, citing an engraved bullet casing as evidence the killing was political. As of August 2026 the judge had not yet ruled whether the case proceeds to trial.

None of these prove that anyone gave an order, which is the whole difficulty. Each shows a person acting alone after long exposure to the idea that a group or a person had to be stopped, and that is hard to prosecute and easy to deny.

The Same Pattern, Four Different Ideologies

A mechanism that only ever described one political side would be a talking point rather than a description of how violence works. These four attackers came from four incompatible worldviews, and each left a documented trail of what they had been consuming.

An anti-abortion militant, an Islamist student, a left-wing Trump opponent, and a neo-Nazi. None was ordered to act, and each had spent months or years being told by voices they trusted exactly who was responsible for what was wrong.

One narrative can also produce more than one attacker. The Great Replacement claim that Bowers cited in 2018 turned up again in the manifesto of the man who killed 10 people at a Buffalo supermarket in May 2022. Neither man knew the other, and they had read the same thing.

Which ideology dominates changes from year to year, so any count is a snapshot. The Center for Strategic and International Studies counted at least five left-wing plots or attacks through July 2025 against one on the far right, the murder of Melissa Hortman, the first time in more than three decades that left-wing incidents led. Over the previous decade the death toll ran the other way, with right-wing attacks killing 112 people and left-wing attacks killing 13.

A caution on reading these cases, which the research community stresses. A case is only strong when the attacker’s own words and their documented media trail point the same way. Where motive is inferred from political affiliation alone the link is guesswork, and lone-actor violence often mixes ideology with private grievance. The four cases above qualify, and many claims made in the days after an attack do not.

What the Peer-Reviewed Research Found

The strongest evidence is a 2023 study in the American Economic Journal by Karsten Müller and Carlo Schwarz, which the Institute for Replication independently reproduced in 2025 and confirmed. It compared more than 3,000 U.S. counties and used Twitter’s early adoption at the 2007 South by Southwest festival as a natural experiment, so the comparison is between places that happened to get Twitter early and places that did not.

A one-standard-deviation increase in Twitter use was associated with a 32% larger rise in anti-Muslim hate crimes after the 2016 primaries. Before 2015 there was no such relationship, which is what makes the timing hard to dismiss.

The chain the authors traced runs in a single direction. Trump’s tweets about Muslims predicted a same-day rise in xenophobic tweets from his followers, which predicted more cable news attention, which predicted more hate crimes in the days that followed.

Elite speech matters more than ordinary speech. Byman’s Brookings analysis found that hateful rhetoric from political leaders is most dangerous when other elites second it, because endorsement from the top tells an audience the prejudice is now sayable.

The pressure shows up in the workload of the officers who guard Congress. Capitol Police count both explicit threats and concerning behavior, and the combined caseload has nearly doubled in two years.

Threat cases against members of Congress rose 87% between 2023 and 2025.

Threat Cases Investigated by U.S. Capitol Police

Category	Value
2023	8008
2024	9474
2025	14938

Source: U.S. Capitol Police, annual threat assessment case figures.

These cases cover members of Congress, their families, and their staff, and the 2025 jump came in the same year as the killings of Melissa Hortman and Charlie Kirk. Capitol Police also tripled their formal agreements with local departments that year, from roughly 115 to more than 350.

One measure moved the other way. Reported hate crimes fell in 2025 to their lowest total in five years, even as threats against officials climbed.

Reported hate crimes and reported threats are two different measures, and in 2025 they moved in opposite directions.

FBI hate crime incidents reported

Period	Value
2024	11,679
2025	10,881
Change	-7%

Hate crime counts depend on which police departments report and on whether victims come forward, so a drop can reflect reporting as much as behavior. Threat caseloads track a different population, which is public officials, and they rose. Neither number is the whole picture.

The rhetoric also travels faster than any correction. Soroush Vosoughi, Deb Roy, and Sinan Aral studied 126,000 stories on Twitter for a 2018 paper in Science and found that false ones were 70% more likely to be retweeted than true ones. True stories took about six times as long to reach 1,500 people.

Why the Law Rarely Reaches the Speaker

Stochastic terrorism is not a crime in U.S. law, and no one has ever been charged with it. The standard for punishing speech that incites violence comes from Brandenburg v. Ohio (1969), under which speech is punishable only if it is intended and likely to produce imminent lawless action. Rhetoric that demonizes a group over weeks or months almost never meets that standard, even when violence follows.

Some legal scholars want that changed, and they say so openly. Liam Pagán argued in the Drexel Law Review in 2024 that Brandenburg should be struck down and replaced with a standard drawn from international human rights law. No court has adopted that view.

Other countries drew the line elsewhere. Germany’s Volksverhetzung law criminalizes incitement to hatred against segments of the population, and the European Union’s Framework Decision on Racism and Xenophobia requires member states to criminalize public incitement to violence or hatred. The United States has no equivalent, which is a constitutional choice rather than an oversight.

The legal gap is not the end of accountability. Officials who dehumanize communities can be censured by colleagues, questioned by constituents, and voted out, and platforms can enforce the rules they already publish. Treating the rhetoric as ordinary political speech is a choice, not a legal requirement.

The Honest Disagreement

Serious people reject this term, and the objections are strong enough that anyone using it should know them. The dispute is not between people who care about violence and people who do not. It is about whether the concept describes anything precise enough to be useful.

The academic case comes from James Angove, whose 2024 article in Critical Studies on Terrorism offers “critical reflections on an emerging concept”. His objections are specific rather than rhetorical.

• The definitions drift. Writers use the phrase for public demonization, for incitement, for the violence itself, and for the unknown people who carry it out. A word doing four jobs does none of them cleanly.
• It gets defined by example. Pointing at cases is easier than setting criteria, which makes the term easy to deploy in an argument and hard to test.
• It may not be terrorism. Most working definitions require someone who commits violence, and this one is applied to a speaker who does not.
• The word “stochastic” oversells the evidence. It borrows the authority of statistics when nobody can produce a probability linking one speech to one attack.
• Anyone can be accused after the fact. Any speaker whose audience contained a violent person becomes a candidate, which is what makes the label easy to misuse.

The free-speech case is blunter. The Foundation for Individual Rights and Expression calls it “an alarmist term used to justify censoring speech that might randomly incite someone to commit some unspecified harmful acts.” Writing at Astral Codex Ten, Scott Alexander argued that the concept gets deployed opportunistically by people who have not noticed that the same argument works against their own side, and that the liberal answer is to allow the criticism, reject the violence, and put the responsibility on the person who pulls the trigger.

Both objections land, and neither is a defense of the rhetoric. Angove does not throw the idea out. He argues for narrowing it to what it describes well, which is authoritarian politics that runs on demonizing and dehumanizing a group rather than any hostile speech that happens to precede violence. That narrower version is the one this page uses, and it is why the test above lists six specific signals rather than a general impression. Amman and Meloy’s framing helps here too, because a threat-assessment concept is a tool for deciding whom to watch, not a basis for prosecuting anyone.

What the Evidence Cannot Prove

The research supports a claim about populations, not about individuals, and the difference is where most bad arguments start.

• No study connects one speech to one attack. The Müller and Schwarz finding compares counties and measures rates. It cannot tell you why any particular person picked up a gun.
• Motive is usually mixed. Lone-actor violence tends to combine ideology with private grievance, job loss, isolation, or untreated illness, and reading motive from party registration alone is guesswork.
• The label carries no legal weight. Calling something stochastic terrorism creates no criminal exposure and no civil liability, so it is an argument about responsibility rather than a finding.
• Anger is not incitement. People are entitled to be furious about policy, and treating fury as violence makes the distinction useless.
• Overuse is the fastest way to lose the argument. Applying the term to any rhetoric you dislike drains it of meaning and gives critics the easiest possible opening.

---

Frequently asked questions

Can anyone be charged with stochastic terrorism? No. No U.S. statute creates the offense, so charges land on the attacker, usually as murder or domestic terrorism. The phrase comes from threat-assessment research rather than from the criminal code.

Do most Americans support political violence? No. UC Davis researchers surveyed more than 8,000 adults and reported in June 2026 that support had stayed largely stable from mid-2024 to mid-2025, with no rise in personal willingness to commit political violence and a slight decline in willingness to kill. The risk comes from a small number of people acting on a widely shared story.

Are online platforms legally responsible when this rhetoric spreads? Generally not. Section 230 of the Communications Decency Act shields platforms from liability for what users post, so most reform proposals ask for algorithmic transparency and enforcement of the platforms’ own published rules rather than new liability.

What you can do

1. Name the technique when you see it, and name it accurately. Say which of the six signals is present rather than reaching for the label. “He has called the same group vermin in nine speeches” persuades a skeptic; “that’s stochastic terrorism” does not.

2. Do not amplify without context. Sharing hostile rhetoric bare spreads it further. If you share it, quote it exactly and say what the speaker is doing.

3. Report threats to the office that investigates them. Threats against members of Congress or their staff go to the U.S. Capitol Police (202-224-0908), whose Threat Assessment Section handles them. Threats against federal judges go to the U.S. Marshals Service. Threats against a local election office go to that office and to your state election board, which routes them to the Justice Department’s election threats task force.

4. Tell your members of Congress to fund hate crime enforcement and require platform transparency. The FBI counted 10,881 hate crime incidents in 2025, and the Marshals logged 564 credible threats against judges in fiscal 2025. Ask for the enforcement money and the algorithmic disclosure using the letter and call script below.

5. Support the groups doing this work. The ADL tracks hate incidents and publishes guidance for families, the Brennan Center follows election misinformation and surveys the officials being threatened, and PEN America offers practical tools for talking to someone who shares it.

If someone you know is showing signs

Look for a pattern, not a single statement. Warning signs include rising hostility, justification of violence, withdrawal from relationships, obsessive media consumption, or a hardening belief that one group is responsible for everything wrong.

• Choose a private, calm moment to talk.
• Lead with concern, not accusation. “I care about you, and I’ve noticed you seem more stressed and more angry lately.”
• Ask what sources they trust and why. Listen before responding.
• Focus on one specific claim at a time, not their entire worldview.
• Keep the relationship open, because isolation makes radicalization worse.
• Set boundaries if the behavior is harming you or others.
• If there is any talk of violence, threats, or self-harm, this is no longer a conversation problem. Call 911 if someone is in immediate danger. Call or text 988 (Suicide and Crisis Lifeline) if someone is in a mental health crisis.

The goal is not to win the argument in one conversation. It is to keep the door open long enough for doubt and reflection to return.