Did Grok tell the truth about right-wing terrorism? On June 18, 2025, Grok said available data suggested right-wing political violence had been more frequent and deadly since 2016; that conclusion is defensible for several lethal-extremism measures, but not as a universal claim about every kind of political violence.
The exchange began when an X user asked Grok whether the political left or right had been more violent since 2016. Grok cited the January 6 Capitol riot and the 2019 El Paso mass shooting. Elon Musk responded that the answer was a “major fail” and said he was “Working on it,” according to contemporaneous reporting by The Independent.
The evidence supports a qualified reading of Grok’s answer, not a blanket verdict that Grok is always truthful or that one political movement is uniformly violent. The central dispute concerns definitions, datasets, time periods, and who controls a commercial AI product’s politically sensitive responses.
Key takeaways
- On June 18, 2025, Grok said available data suggested right-wing political violence had been more frequent and deadly than left-wing political violence since 2016.
- Elon Musk called Grok’s answer a “major fail” and said he was “Working on it,” but the contemporaneous report did not provide a competing dataset.
- According to the ADL Center on Extremism’s 2025 report covering 2024, all 13 identified extremist-related killings in the United States in 2024 were connected to right-wing extremists.
- According to the same ADL report, far-right extremists were responsible for 328 of 429 extremist-related killings over the preceding decade, or 76 percent.
- ADL, START, the FBI, and DHS use different definitions and measures, so extremist murders, terrorist attacks, riots, threats, and broader political violence cannot be treated as interchangeable.
What happened between Grok and Elon Musk?
The dispute began on June 18, 2025, when an X user asked Grok whether the political left or right had been more violent since 2016. Grok answered that available data suggested right-wing political violence had been more frequent and deadly, citing the January 6 Capitol riot and the 2019 El Paso mass shooting as examples. The exchange was reported by The Independent.
Free tools Windows power users keep installed
One-click scans. No signup required.
Grok’s reported answer was: “Since 2016, data suggests right-wing political violence has been more frequent and deadly, with incidents like the January 6 Capitol riot and mass shootings (e.g., El Paso 2019) causing significant fatalities.”
Musk rejected that conclusion publicly, writing: “Major fail, as this is objectively false.” Musk added, “Working on it.” Those statements establish that Musk disputed the answer and indicated that changes might be considered; they do not, by themselves, establish which dataset Musk was using or prove that he changed Grok’s instructions.
The headline framing that Grok was “telling the truth” is an editorial characterization, not evidence that Grok is generally reliable. The more precise question is whether Grok’s particular claim is defensible under particular definitions, time periods, and measures of violence.
Was Grok answering about terrorism or political violence?
Grok’s quoted answer referred to “right-wing political violence,” not to a narrowly defined statistical category called right-wing terrorism. That distinction matters because terrorism is one form of violence, while political violence is a broader and less standardized category.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Fix the driver behind crashes, sound loss and screen glitches3Clear out junk files and repair common Windows errorsSTART, the University of Maryland’s National Consortium for the Study of Terrorism and Responses to Terrorism, defines terrorism as the threatened or actual use of illegal force and violence by a non-state actor to pursue a political, economic, religious, or social goal through fear, coercion, or intimidation. START program manager Erin Miller cautioned: “It’s important to understand that terrorism is just one type of violence, even within the broader landscape of ideologically motivated attacks.” Read START’s 2017 analysis and definition of terrorism.
Federal terminology is also not perfectly uniform. The U.S. Government Accountability Office reports that the FBI and Department of Homeland Security treat a domestic-terrorism incident as involving the use or threat of force or violence, the potential for a federal violation, and an apparent intent to further social or political goals. GAO also notes that agency definitions are not identical and that incidents rooted in state or local criminal activity may go unreported. The GAO report on the federal domestic-terrorism strategy explains those limitations.
Consequently, a claim about right-wing terrorism cannot automatically answer a question about every politically motivated assault, protest confrontation, riot, property attack, threat, or act of intimidation. A responsible comparison must name the category being measured.
What do the ADL figures show about lethal extremist violence?
The ADL’s recent data support a qualified version of Grok’s underlying conclusion: lethal extremist violence in the United States has been heavily concentrated in far-right or right-wing extremist categories in the measures ADL reports.
According to the ADL Center on Extremism’s 2025 report covering 2024, domestic extremists killed 13 people in the United States in 2024, and all 13 identified extremist-related killings were connected to right-wing extremists. According to the same report, far-right extremists were responsible for 328 of 429 extremist-related killings over the preceding decade—76 percent of the total.
The ADL report also identifies the 2019 El Paso Walmart attack as a far-right white-supremacist mass shooting in which 23 people were killed. Large attacks such as El Paso matter because a small number of mass-casualty events can substantially affect a comparison of fatalities over a particular period.
ADL does not present those figures as a measurement of every kind of political violence. The ADL dataset tracks extremist-related murders across categories including right-wing, left-wing, domestic Islamist, and other extremism. The dataset does not include every plot, threat, hate crime, armed standoff, or nonfatal incident. ADL also says its 2024 totals are preliminary and may change; yearly totals typically stabilize only after about five years.
These limitations do not make the ADL evidence irrelevant. They define what the evidence can support. The ADL figures are strong evidence about identified extremist-related killings, especially the distribution of lethal far-right extremist violence. They are not a complete scoreboard for all political conflict or all violence associated with people on the political left and right.
What does START data show, and where does it stop?
START provides a broader historical terrorism dataset, but the cited analysis ends in 2016 and therefore cannot independently settle a question about the entire period from 2016 through the present.
According to START’s 2017 analysis, the Global Terrorism Database recorded 2,794 terrorist attacks and 3,659 resulting deaths in the United States from 1970 through 2016. During the first seven years of the 2010s compared with the 2000s, the proportion of U.S. terrorist attacks attributed to right-wing extremists rose from 6 percent to 35 percent, while the proportion attributed to left-wing or environmental extremists fell from 64 percent to 12 percent. Those figures are reported in START’s November 2, 2017 report.
START’s figures measure terrorist attacks, not every incident of political violence. Ideological classification can include more than one motivation, and classifying an attack does not characterize an entire political population or movement. START’s Erin Miller stated: “This report is not intended to be a comprehensive threat assessment or to characterize the behavior of entire ideological movements.”
The end date is equally important. A dataset ending in 2016 can provide historical context for Grok’s reference to trends since 2016, but it cannot serve as a complete measure of events after 2016. Using START’s 1970–2016 figures as if they were a current 2016-to-2025 comparison would overstate what the research establishes.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
How do the main political-violence datasets differ?
The datasets differ in their definitions, time windows, outcomes, and attribution rules. The following comparison shows why the numbers should not be combined into one undifferentiated claim.
| Source | Time period | Primary measure | What the cited evidence says | Important limitation |
|---|---|---|---|---|
| ADL Center on Extremism, 2025 report covering 2024 | 2024 | Identified extremist-related killings | According to ADL, 13 people were killed by domestic extremists; all 13 identified killings were connected to right-wing extremists. | The total is preliminary and does not include every plot, threat, hate crime, standoff, or nonfatal incident. |
| ADL Center on Extremism, 2025 report covering 2024 | The preceding decade in the report | Extremist-related killings | According to ADL, far-right extremists were responsible for 328 of 429 killings, or 76 percent. | Mass-casualty events can strongly influence long-term fatality totals, and the category is not identical to all political violence. |
| START, 2017 analysis | 1970–2016 | Terrorist attacks and resulting deaths | According to START, the database contained 2,794 attacks and 3,659 deaths overall; the right-wing share of attacks rose from 6 percent in the 2000s to 35 percent in the first seven years of the 2010s. | The analysis ends in 2016, covers terrorism rather than all political violence, and can assign more than one ideological motivation. |
| FBI and DHS definitions summarized by GAO | Federal domestic-terrorism framework | Use or threat of force or violence connected to political or social goals | GAO describes the elements used by the agencies to identify domestic-terrorism incidents. | FBI and DHS definitions are not identical, and some incidents rooted in state or local criminal activity may be unreported. |
Is right-wing political violence more common than left-wing violence since 2016?
The most defensible answer from the supplied evidence is yes for the cited measures of lethal far-right extremist violence, but the evidence does not establish a universal yes for every form of political violence since 2016.
The ADL’s 2024 findings and its preceding-decade figures show a strong concentration of extremist-related killings in the far-right category. Those findings make Grok’s general direction defensible if Grok was primarily summarizing lethal extremist violence. START’s historical analysis also recorded a rising right-wing share of U.S. terrorist attacks during the first seven years of the 2010s, although that analysis does not cover the full period after 2016.
The phrase “more frequent and deadly” contains two separate claims. Frequency asks how many incidents occurred; lethality asks how many people were killed. A movement could account for more incidents but fewer deaths, or fewer incidents but more deaths if its attacks were more lethal. The cited dossier supplies meaningful evidence on extremist-related killings and terrorist attacks, but it does not provide one comprehensive, directly comparable dataset covering every kind of left- and right-wing political violence from 2016 onward.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallOutdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchThe conclusion therefore should not be expanded into claims that every conservative, Republican, right-leaning voter, or MAGA supporter is violent. The conclusion also should not treat every left-wing protest, riot, property attack, or political confrontation as terrorism. Attribution belongs to specific actors and incidents, not automatically to everyone who shares a broad political identity.
Why did Musk call Grok’s answer a “major fail”?
Musk called the answer a “major fail” because he judged Grok’s conclusion to be “objectively false.” The contemporaneous report records Musk’s rejection and his statement that he was “Working on it,” but it does not present a competing dataset or explain which definition of violence he believed Grok should have used.
There are several methodological reasons someone might object to Grok’s wording without disproving the underlying evidence. Grok used the broad phrase “political violence,” while the strongest cited support comes from narrower categories such as extremist-related murders and terrorism. A critic could also dispute the time window, the treatment of incidents with mixed motivations, the choice between incident counts and fatalities, or the inclusion of events such as January 6 and El Paso.
Those are legitimate questions for a data comparison. They are not answered by a bare assertion that the result is false. Musk’s public response is evidence of a dispute over Grok’s output and its political framing; it is not a substitute for a source, definition, or reproducible analysis.
Does one answer prove that Grok is politically biased?
No. One politically sensitive answer, even an answer that appears to favor one side of a comparison, does not prove that Grok is generally politically biased or generally truthful.
A model’s response can depend on the wording of the prompt, the model version, retrieved information, system instructions, training data, safety rules, post-training adjustments, and the way the model weighs conflicting sources. A fair bias assessment would need repeated tests using balanced prompts, a fixed model version, transparent source review, and comparisons across multiple political topics and factual questions.
The dispute does show why a commercial chatbot should not be treated as an independent authority. Grok’s output is a product behavior that can be evaluated, challenged, and potentially changed. The output is not itself a primary dataset on political violence.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How does Grok work as a commercial AI product?
xAI describes Grok as a consumer assistant available through the web and mobile applications, with features including chat, multimodal generation, voice interaction, file analysis, connectors, and free and paid usage tiers. The official Grok product documentation describes the product’s capabilities and access model.
xAI’s Grok 4 model card says the training recipe includes publicly available internet data, third-party data, user or contractor data, and internally generated data. The model card also describes filtering, supervised fine-tuning, human feedback, verifiable rewards, and model grading, and says xAI publishes consumer system prompts. The Grok 4 model card documents those development layers.
Those disclosures explain why a model output can reflect multiple layers of data and instruction. They do not establish why this particular answer was generated, whether the answer used a specific source, or whether Musk personally changed a system prompt. A product description and model card are not evidence that Grok’s political conclusions are correct or incorrect.
Readers who want a practical introduction can look for a current Grok AI guide book, but publication date and version matter because AI interfaces and capabilities change. A guidebook may explain how to use the product; it should not be treated as evidence about the accuracy of Grok’s political answers or as an official xAI publication.
Did Musk try to change Grok’s answer?
The available evidence shows that Musk said he was “Working on it,” but it does not prove that he completed a change to Grok’s answer, system prompt, training data, retrieval system, or model weights.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →Best Value
That distinction is important. A product owner can comment publicly on an output without that comment showing what technical change, if any, followed. Establishing a change would require primary evidence such as a documented system-prompt revision, a model-version update, an official xAI explanation, or reproducible before-and-after testing under the same conditions.
The episode nevertheless raises a legitimate governance question: how should users evaluate a truth-seeking claim when the company owner publicly rejects the model’s political answer? The answer requires transparency about model versions, system instructions, sources, evaluation methods, and post-release changes—not assumptions about what happened behind the scenes.
How should readers evaluate claims about political violence?
Readers can avoid most misleading comparisons by asking six questions before accepting a claim about left-wing or right-wing violence.
- What is the definition? Determine whether the claim concerns terrorism, extremist murder, domestic terrorism, riot, threat, property damage, assault, or a broader category called political violence.
- What is the time period? Check whether the comparison begins in 2016, ends in 2024, uses the preceding decade, or relies on START’s historical period ending in 2016.
- What is being counted? Incident totals, fatalities, injuries, plots, arrests, and property damage answer different questions. A fatality comparison cannot be presented as an incident comparison.
- How was attribution made? Ask whether the label applies to an individual perpetrator, an organized group, an ideological motivation, or a broad political movement. Mixed motivations require particular caution.
- Which source produced the number? ADL, START, FBI, DHS, GAO, and other organizations may define and classify incidents differently. A number should stay attached to the dataset that produced it.
- Is the figure final? Preliminary totals can be revised, and ADL notes that yearly extremist-murder totals may take about five years to stabilize. A current-looking number may not be a final number.
This checklist also helps with AI answers. Ask the chatbot to name the dataset, define the category, separate attacks from deaths, state the exact dates, identify preliminary figures, and provide links to primary research. If the chatbot cannot do those things, its confident wording should not be mistaken for a settled finding.
Recommended Free Tools
What should readers conclude?
Musk rejected Grok’s answer, but Musk’s rejection did not disprove it. The available evidence makes Grok’s underlying direction defensible for specific measures of lethal far-right extremist violence, including ADL’s 2024 findings and its preceding-decade murder data.
The careful conclusion is narrower than either side’s slogan: right-wing extremist violence accounts for a substantial share of documented lethal extremist violence in the cited datasets, but no single figure here measures every form of political violence since 2016. The real issue is the interaction of definitions, datasets, model behavior, and the power to alter a commercial AI product’s responses.
Frequently Asked Questions
Did Elon Musk prove that Grok’s answer was false?
No. Musk publicly called Grok’s answer “objectively false,” but the contemporaneous report did not provide a competing dataset or definition that disproved Grok’s claim. The available ADL evidence supports a qualified version of Grok’s conclusion about lethal far-right extremist violence.
What do the ADL statistics say about right-wing extremist violence?
ADL’s cited figures show that all 13 identified extremist-related killings in the United States in 2024 were connected to right-wing extremists, while far-right extremists accounted for 328 of 429 extremist-related killings over the preceding decade, or 76 percent. ADL says those figures are preliminary and do not measure every form of political violence.
The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Does Grok’s answer prove that Grok is politically biased?
No. A single answer does not establish that Grok is generally politically biased. A reliable bias assessment would require repeated tests with balanced prompts, a fixed model version, transparent source review, and comparisons across multiple political topics.
Did Musk personally change Grok’s answer or system prompt?
The available evidence shows only that Musk said he was “Working on it.” The evidence does not prove that he personally changed Grok’s system prompt, training data, retrieval system, model weights, or final answer.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
