Projects · Research
Government Hate Speech
When those who hold public power vilify a group of people, the law in most places has almost nothing to say. This project asks why, and what the law should do about it.
I · The project
Words spoken with the authority of the state
A minister calls a minority “vermin.” A head of state tells a crowd that a religious community is a fifth column. A governor describes migrants as an invasion. Unless such words cross the narrow line of incitement to imminent violence or genocide, almost no legal system treats them as a wrong. This project calls the phenomenon government hate speech: the use of speech in the political arena, by a person who holds or seeks public power, to vilify, dehumanize, or incite hatred against a group defined by shared traits such as race, ethnicity, religion, nationality, gender, or sexual orientation.
The definition reaches the ministerial statement and the official social media account, and also the campaign rally and the party broadcast, because politicians and parties are the vehicles through which state power is sought and used. It does not reach criticism of policies, ideas, or the conduct of individuals.
The project’s working paper, Words of the Sovereign: The Philosophical Foundations of Government Hate Speech as a Crime, argues that the law’s silence is no accident. It traces it to three assumptions in modern legal thought: that words, unlike acts, cause no real harm; that the state needs an unfettered province of action and is restrained only for the benefit of individual citizens; and that law is an exercise of pure reason, detached from emotion and narrative. In their place the paper proposes that the state is a speaker, that it can do wrong by speaking, and that official words are acts of authority. Government hate speech is an abuse of public power, and its harm comes from the authority of the speaker, not only from the content of the words.
The paper draws its evidence deliberately from Asia, Africa, Latin America, Europe, and North America, and ends by sketching the elements of a distinct crime, the institutional answers to the problem of a state asked to prosecute itself, and replies to the strongest objections.
Working paper in draft (2026). It develops an earlier working paper, “A Case for Government Hate Speech.”
II · Statistics
A rising tide
No global dataset measures government hate speech as such. The figures below come from national monitoring projects, cross-national indices, and peer-reviewed studies, and are indicators rather than a single measure.
hate speech events against religious minorities recorded in India in 2025, up from 668 in 2023
India Hate Lab
of India’s 2024 events were delivered by politicians, 452 of them by leaders of the governing party
India Hate Lab
countries and territories where governments harassed religious groups in 2022
Pew Research Center
perpetrators of the 1994 Rwandan genocide, about 10% of the violence, attributed to RTLM radio broadcasts
Yanagizawa-Drott
Recorded hate speech events in India
In-person events targeting religious minorities, all speakers
Who delivered India’s 2024 events
1,165 events split by type of speaker
Most frequent speakers who held office
Show the figures as a table
| Indicator | Value | Speaker measured | Source |
|---|---|---|---|
| India hate speech events, 2023 / 2024 / 2025 | 668 / 1,165 / 1,318 | All speakers | India Hate Lab |
| Events by politicians, 2024 | 462 (452 by governing-party leaders), up from 100 in 2023 | Politicians | India Hate Lab |
| Events with dangerous speech, 2024 | 259 (22.2%) | All speakers | India Hate Lab |
| Events with explicit calls for violence, 2025 | 308 | All speakers | India Hate Lab |
| Government Restrictions Index, global median | 1.8 (2007) to 3.0 (2022) | Governments as institutions | Pew Research Center |
| Countries with government harassment of religious groups, 2022 | 186 of 198 | Governments as institutions | Pew Research Center |
| Murders of Indigenous people in Brazil, 2019 / 2020 | 113 / 182; invasions of Indigenous territories up 137% | Violence after official rhetoric (head of state) | CIMI |
| Share of Rwandan genocide violence attributed to RTLM | About 10% (≈51,000 perpetrators) | State-aligned broadcaster | Yanagizawa-Drott |
From words to violence
Statistical studies consistently find an effect. In the United States, a head of state’s tweets about Muslims predicted rises in anti-Muslim hate crimes in the following days. In Germany, anti-refugee posts by a populist party predicted violent crimes against refugees. Experiments show that inflammatory speech by a leading candidate emboldens people who already hold prejudices to act on them, especially when other leaders fail to condemn it. The relationship is one of contribution, not sole causation: official words reliably accompany violence, often precede it, and in several documented cases measurably increase it. The paper is open about the limits of this evidence: few studies isolate sitting office-holders, and much of the world collects no relevant data.
III · The record
Official words and communal violence, 1992–2024
A selection of documented episodes in which hateful speech by officials or state-recognized leaders preceded or accompanied violence against the group targeted.
- Serbia · 1992
A party leader tells a rally in Hrtkovci that the village’s Croats must leave; many do, under threat. He is later convicted of instigating deportation and persecution as crimes against humanity.
- Rwanda · 1992
A senior ruling-party official calls Tutsi “cockroaches” in a speech at Kabaya, two years before the genocide.
- Kenya · 2007–08
A national commission documents ethnic incitement by sitting MPs and candidates around the election. 1,133 people are killed.
- South Africa · 2015
The Zulu king, whose office is state-recognized and funded, tells foreigners to “pack their bags and go home.” Attacks the next month leave at least seven dead and thousands displaced.
- Myanmar · 2017
The armed forces commander describes operations in Rakhine State as finishing “the unfinished job.” More than 700,000 Rohingya flee to Bangladesh.
- Indonesia · 2019
Soldiers and militia call Papuan students in Surabaya “monkeys.” Protests spread to some thirty cities, with riots in Manokwari and Sorong.
- India · 2020
A junior minister leads a campaign crowd chanting that “traitors” should be shot. Riots in north-east Delhi the next month leave 53 dead, most of them Muslim.
- Ethiopia · 2021
During the war in Tigray, the prime minister urges citizens to “bury” Tigrayan forces; the post is removed for inciting violence.
- United States · 2024
False claims by a former president and a senator that Haitian immigrants in Springfield, Ohio, were eating pets are followed within a week by more than thirty bomb threats.
IV · Normative frameworks
Six responses, no strategy
The law reaches official speech when it becomes genocidal, and reaches hateful speech when private persons utter it. It does not reach the large territory in between.
| Response | Instruments | What it reaches | What it leaves out |
|---|---|---|---|
| International criminal law | Genocide Convention art. III(c) and art. IV; Rome Statute art. 25(3)(e); ICTR Akayesu and Nahimana (Media case); IRMCT Šešelj | Direct and public incitement to genocide, including by “constitutionally responsible rulers”; speech forming part of a widespread or systematic attack | Everything short of mass atrocity. The ICTR Appeals Chamber declined to hold that hate speech alone amounts to persecution. |
| Treaty duties of states | ICERD art. 4(c); ICCPR art. 20(2) | States must not let public authorities promote or incite racial discrimination, and must prohibit advocacy of hatred that incites discrimination, hostility or violence | No individual liability. ICERD art. 4(c), the only provision aimed at the state as speaker, covers race only. ICCPR art. 20(2) addresses the state as regulator. |
| Soft law | Rabat Plan of Action (2012); UN Strategy and Plan of Action on Hate Speech (2019); PACE Resolution 2275 (2019); CoE CM/Rec(2022)16 | Urges political leaders to refrain from and condemn hate speech; treats the speaker’s status as a factor; encourages party codes of conduct | Non-binding; treats criminal law as a “last resort”; relies on parties and parliaments to discipline their own members. |
| General hate speech offenses | Canada Criminal Code s. 319; UK Public Order Act 1986; India BNS s. 196; Indonesia EIT Law art. 28(2); Kenya NCI Act s. 13; South Africa Act 16 of 2023 | Applies to officials as to anyone else | Official status is at most a sentencing or contextual factor; rarely enforced against sitting office-holders. |
| Electoral law | India Representation of the People Act 1951 s. 123(3), (3A); Indonesia Law No. 7 of 2017 art. 280(1)(c) | Appeals to religion or enmity, or group insults, by candidates during campaigns; can void an election | Only campaign speech and only candidates. The rule falls away once the speaker takes office. |
| Parliamentary and party discipline | Parliamentary codes of conduct; party charters | Internal sanctions against members | Depends on the majority’s willingness; legislative privilege shields speech in the chamber. |
Constitutional texts
Constitutions that mention hate speech are recent: Brazil (1988), Slovenia (1991), South Africa (1996), Venezuela (1999) and Ecuador (2008). Each uses it to narrow the individual’s freedom of expression or to direct the state to punish others. None treats the state as a possible author of hate speech.
The United States
Private speech is protected unless it is directed to and likely to produce imminent lawless action (Brandenburg). Government speech falls outside the Free Speech Clause altogether (Summum), so it is neither protected nor constrained. Courts have used officials’ hostile words only as evidence that a decision was tainted (Masterpiece Cakeshop), and looked past them in Trump v. Hawaii.
The closest case
In Féret v. Belgium (2009), the European Court of Human Rights upheld ten years’ ineligibility for office against a parliamentarian convicted of incitement. The sentence rested on a general offense; the speaker’s status justified the restriction but was not the foundation of the wrong. In Mugesera (2005), the Supreme Court of Canada upheld a deportation order based on the Kabaya speech.
V · The proposal
Elements of a proposed offense
Offered as a starting point for doctrinal development, not as a finished statutory text.
Actor
Holders of public office at any level, members of the state apparatus, and politicians, including candidates and party members, speaking in an official capacity or in the political arena. The closer to office, the graver the offense.
Conduct
Public expression that, in context, vilifies or dehumanizes a group defined by a protected characteristic, or advocates hatred of it or its exclusion from the political community. No proof of imminent violence or resulting harm is required.
Mental element
Intent to vilify or dehumanize, or knowledge that the expression would be understood that way. Negligent or careless speech falls outside.
Context
Assessed with the Rabat factors: social and political context, the speaker’s position and authority, content and form, reach, and likelihood of contributing to discrimination, hostility, or violence. Imminence is not required.
Exclusions
Good-faith criticism of the policies or conduct of groups or their leaders, statements of fact in the course of official duties, and genuine public debate on immigration, security, or religious policy that does not vilify a group as such.
Consequences
Graduated penalties, with disqualification from public office at their center. Custodial sentences are reserved for the most serious and persistent cases.
The enforcement paradox
A crime of government hate speech names the state as perpetrator while depending on the state to enforce it, and international law, which does not rest on coercive power, compounds the problem rather than solving it. The paper argues that the answer lies in institutional design and in the expressive force of law: law persuades as well as compels.
VI · Testing the argument
What the data says, claim by claim
The figures below are computed live from the country data in the explorer. They show association across countries, not cause, and some point against the argument as well as for it.
Evidence still to gather
These sources would test the claims the cross-country data cannot reach. If you work with any of them and want to collaborate, get in touch.
- Is the link causal? Re-run the event studies in the replication files of Yanagizawa-Drott (RTLM radio), Müller and Schwarz (social media) and Edwards and Rushin (hate crimes), with data up to 2025.
- Who is prosecuted? Code the cases in Columbia’s Global Freedom of Expression database and the European Court of Human Rights’ HUDOC by whether the speaker was an official, a politician or a critic of the government.
- Does hate pay? Match Manifesto Project and Global Party Survey positions on minorities to election results.
- Harm short of violence. Track minorities’ sense of belonging and public tolerance (EU-MIDIS, World Values Survey, Afrobarometer) in years when officials speak against them.
- Before and after, district by district. Indonesia’s SNPK violence data (1997–2014) and India’s riot records allow speech-by-speech checks below the national level.
VII · Explore the data
Do your own analysis
Filter the documented episodes and court decisions, compare the legal responses against the criteria that matter to you, test a statement against the proposed offense, and download the data.
| Year | Country | Region | Speaker | What was said | What followed | Legal follow-up |
|---|
Speaker categories follow the working paper: Office-holder or state agent is the core of the definition, Politician or candidate the outer ring, and State-recognized authority lies at its edge.
Recorded violence in each episode country in the 12 months before and after the month of the statement, which is marked in dark red. The charts show what happened around each statement, not whether the statement caused it.
UCDP counts only organized violence that kills at least 25 people a year in the same conflict, and attacks by one group on unarmed civilians. It misses most riots, so a flat line does not mean nothing happened. For Serbia, UCDP places 1992 fighting in Croatia and Bosnia under those countries, so the Serbian series falls to zero. ACLED records every reported event, including riots and mob violence, but starts later in many countries (India 2016, Indonesia 2015, Myanmar 2010, United States 2020), so earlier months show as no data. Sources: UCDP GED and Non-State 26.1; ACLED weekly regional data to 17 October 2026 (riots, mob violence) and ACLED country-month data from HDX, downloaded 30 October 2026 (other ACLED measures).
Summaries are IMR’s, drawn from the working paper. Links go to the court’s own site or a free legal database.
Tick the criteria you care about. Each legal response gets a score: 2 points where it fully meets a criterion, 1 where it partly does.
Ratings are IMR’s reading of the working paper’s survey (Table 1), not a legal opinion. Hover or tap a cell for the reason.
Think of a specific statement. Answer each question to see whether it would fall within the offense proposed in the working paper.
This is an educational tool based on a proposal for a new offense. It is not legal advice and does not describe the law of any country.
Test the questions yourself with cross-country data: how often political parties use hate speech, how much governments spread false information, restrictions on religion and social hostility, and violence against civilians. Every answer sets out the null and alternative hypotheses, the test used, the p-value and whether the null hypothesis is rejected at the significance level you choose, with a 95% interval and a plain reading.
The country data is being prepared and will appear here.
Nine documented episodes, 1992–2024, with speaker category and outcome.
Twenty international, regional and national decisions with summaries and links.
India Hate Lab, Pew, CIMI and RTLM indicators shown on this page, with sources.
The six legal responses with their ratings on every criterion.
The criteria, ratings, speaker categories and every country variable with its source.
Download CSVAll of the above in one JSON file.
Source: IMR working paper, Words of the Sovereign (draft, 2026), and the sources it cites. Please cite the Institute for Migrant Rights when you use this data.
Report an incident
Heard an official spread hate?
Tell us when a minister, official, police officer, judge or politician speaks against people because of their religion, ethnicity, caste, nationality or migration status. We check each report and add confirmed incidents to the record. You do not have to give your name.
Partners
Working with
Cornell Farmworker Legal Assistance Clinic
Cornell Law School · Ithaca, New York