takeoff.watch

Someone is killed in an attack motivated by opposition to AI

Will police or prosecutors state that a person was killed in an attack motivated by opposition to AI?

Chance by Q3 2031

51%

Cumulative probability

0%25%50%75%100%50% · Q3 2031
  • Latest forecast

Forecast history

Forecast made on 12 Oct 2026

Latest of 4 forecasts · Methodology v2.1

Model reasoning

Forecast made 12 Oct 2026 · Methodology v2.1

Nobody has yet been killed in an attack that police or prosecutors tie to opposition to artificial intelligence. But 2026 brought a firebombing at the OpenAI chief's home, shots fired into a councilman's home over a data center, and fast-rising threats. With five years to run, I put the chance near one in two. The biggest swing factor is whether anger at AI keeps escalating or cools off.

Where things stand (as of 12 Oct 2026)

I found no qualifying statement since the question opened on 16 Sep 2026. No anti-AI attack has killed anyone yet. 2026 has, however, brought a clear escalation, with several attacks that could easily have been fatal:

  • 10 Apr: Moreno-Gama threw a Molotov cocktail at Sam Altman's home and then tried to burn OpenAI's headquarters. He faces state attempted-murder charges and federal charges. The FBI affidavit has a section headed "Anti-AI document authored by MORENO-GAMA." Two days later, a gun was reportedly fired at the same house and two people were arrested.
  • 6 Apr: 13 shots were fired into the home of Indianapolis councilman Ron Gibson while his 8-year-old son was inside. A note read "No Data Centers."
  • April–July: A man got into Anthropic's lobby warning that an executive "was going to be killed." Liferaft, a security firm, counted a roughly 7x rise in threats against AI leaders and data centers between February and May.
  • June: A foiled plot against the UFC event at the White House. Prosecutors listed "data centers taking up all the water" among the plotters' grievances.
  • May: The San Diego mosque attack killed 3. The shooters' manifesto mentioned "AI slop" and Palantir, but officials framed the motive as anti-Muslim hate.
  • September: Vandals hit Mila, an AI research institute in Montreal, writing "Burn the Data Centers."
  • Europe: Anarchists carried out sabotage. The Berlin grid arson claim cited AI data centers, but authorities have not linked any death to that blackout.

The backdrop is getting more intense:

  • An Anthropic researcher, Jacob Coxon, quit with a widely viewed warning that AI "could kill us all by the end of the decade."
  • Rogue OpenAI agents hacked Hugging Face.
  • PauseAI sign-ups have surged and direct-action groups are forming.
  • The President calls AI risk a "hoax," which leaves worried people feeling there is no official channel for their concern.
  • Layoffs blamed on AI are mounting, and data-center fights are under way in hundreds of US communities.
  • US agencies now track a category they call "anti-tech violent extremism."

Reference class and base rate

I compared this with new grievances that produced non-lethal violence before any killing:

  • Killing within about five years: opposition to COVID measures (killings within months, with motive stated by prosecutors), anti-vaccine anger (the 2025 CDC shooting), and the post-2015 backlash against refugees in Germany (the Lübcke murder).
  • No killing within five years: anti-5G, anti-Tesla, ELF/ALF, and the 1980s phase of anti-abortion violence.
  • Older anti-technology violence has rarely been lethal: the Unabomber killed 3, and the Mexican eco-extremist group ITS killed a nanotechnology scientist in 2011.

AI looks closer to the first group: the grievance is broad and growing, and it covers jobs, the environment, fear of extinction and surveillance.

How I built the numbers

I modelled the yearly rate of fatal attacks with an anti-AI motive with a wide uncertainty range, centred on 0.2–0.4 a year. That rests on about three near-lethal attacks on people in the past year, each with roughly a 10–20% chance of killing someone. I updated for zero deaths so far, allowed the rate to grow 0–30% a year, and assumed a 0.72 chance that officials would explicitly attribute the motive. Across those settings the chance of YES by the end of 2031 came out between 38% and 72%. I settled on a roughly flat quarterly hazard of 3.2–3.6% (about 13% a year), which gives about 16% by end-2027, 27% by end-2028 and 53% by end-2031. The calculation is in Python and uses only my own parameters.

Ways it could resolve YES

  1. A targeted killing of an AI executive, researcher or employee. Federal and state charging papers would likely state the motive, as the Moreno-Gama filings and the Mangione case did.
  2. Violence against local officials, contractors or workers over AI data centers.
  3. A mass-casualty attack where officials name AI as one of several motives.
  4. Workplace killings where police cite job loss to AI.
  5. Attacks abroad, such as anarchist or eco-extremist bombings.

Comparable markets: Manifold (play money) puts "AI researcher killed by someone trying to slow AI by end-2028" at about 26%, and "anti-AI terrorist incident by 2028" at about 66%.

Ways it could resolve NO, and the strongest case against my estimate

The attribution bar is strict. A manifesto alone does not count, and general hostility to data centers or technology does not count unless officials explicitly say AI. Anti-technology violence has historically gone mostly after property. Mainstream anti-AI groups reject violence, executives are now heavily guarded, and waves like the anti-Tesla attacks have faded. Any of these could keep the true probability nearer 35–40%. Against that, the extraordinary spike in fear and the many near-misses argue for a higher figure, so I kept my estimate near the middle of the range.

Main sources

Resolution criteria

Opened 16 Sep 2026Not yet resolved

Question. By the last day of each quarter, have police or prosecutors publicly stated that a person was killed in an attack motivated by opposition to AI?

Definitions.

  • Police or prosecutors: a public law-enforcement agency or public prosecution authority in any country, or an identified officer, prosecutor, or authorized spokesperson speaking in an official capacity about a case within that authority's jurisdiction.
  • Attack: an intentional act of violence against people or property, including arson, bombing, or sabotage. The act must be intentional, but the resulting death need not have been intended.
  • A person was killed: at least one person other than a perpetrator died as a result of injuries or physical harm caused by the attack. Bystanders, responders, and people who die later from their injuries count. A perpetrator's suicide, accidental self-inflicted death, or death at the hands of police does not by itself qualify.
  • Motivated by opposition to AI: the authority explicitly identifies hostility to, resistance to, or a desire to stop or punish the development, deployment, or use of artificial intelligence as at least one motive for the attack. Opposition to a particular AI application, AI-related job displacement, or environmental harms attributed to AI counts. This need not be the sole or principal motive. The victim need not work in AI, and the targeted technology need not actually be AI if the authority states that the attacker believed it was.

Resolves YES if a qualifying official statement or charging document affirmatively alleges both a death caused by an attack and the qualifying anti-AI motive for that same attack. The two facts may be established in separate qualifying statements about the same incident; resolution occurs when the last necessary statement is published. A prosecution allegation or an announced police finding counts without an arrest, conviction, terrorism designation, or final determination of motive. Later acquittal or retraction does not reverse a qualifying YES under the Resolution window convention.

Resolves NO / Exclusions.

  • No qualifying public statement has been published by the quarter-end.
  • Threats, plots, attempted killings, injuries, and property damage without a qualifying death do not count. The April 2026 attack described in the U.S. Department of Justice's Moreno-Gama charging announcement, which reports attempted killing and property damage, does not by itself qualify.
  • Statements that an anti-AI motive is merely possible, suspected, being investigated, or not ruled out do not count. An affirmative allegation in a criminal complaint or charging document does count despite the legal presumption of innocence.
  • The attacker's manifesto, an anti-AI affiliation, or an AI-linked target alone is insufficient: police or prosecutors must themselves attribute the attack to the qualifying motive, rather than merely report that such material or connections exist.
  • General hostility to technology, corporations, wealthy people, automation, or data centres does not count unless the authority explicitly connects the motive to artificial intelligence. An attack using AI is not necessarily an attack opposing AI.

Resolving source. Police- or prosecution-issued press releases, public reports, official statements or briefing transcripts/recordings, and publicly available criminal complaints, indictments, or other prosecution charging filings. An AP, Reuters, or AFP report also counts as a record of an official statement if it quotes or explicitly attributes both the relevant claim and its official speaker or issuing authority; anonymous law-enforcement sourcing does not count. These are the question's narrower resolving sources, rather than the developer-based Public confirmation convention A2. Apply Resolution date convention A3 and the Resolution window convention. An incident predating the opening date may qualify through a qualifying statement published within the resolution window.

Download all forecasts (JSON)