Guyana's Ethnic Relations Commission issued 48 cautionary warnings for ethnic hate speech on social media between 29 July and 24 August 2026, the highest single-cycle total the Commission's Media Monitoring Unit has reported this year. That is up from 37 warnings the cycle before it. Standalone racial slurs accounted for 18 of the 48, more than any other category, followed by 12 racially prejudiced remarks and 9 racially derogatory statements.
Guyana already uses artificial intelligence to fight this problem. Since October 2024, the ERC has run eMonitor+, a UNDP-backed AI system that scans Facebook, Instagram, X and YouTube for racist and xenophobic content, and it is a large part of why the Commission's warning numbers have climbed rather than fallen this year. The Commission itself credits "expanded monitoring capacity and more systematic coverage of digital platforms" for the increase.
Whether that is good news depends on which half of the sentence gets the emphasis. Catching more hate speech is not the same as reducing it, and the part of the ERC's operation built to act on what the AI finds, its separate Investigative Unit, is still working through a caseload that predates the tool by two years.
A Record Month, and a Pattern Years in the Making
The 48 warnings issued in the late July to late August cycle break down into seven categories: 18 standalone racial or ethnic slurs, 12 racially prejudiced remarks, 9 racially derogatory statements, 4 racially hateful remarks, 2 racially divisive statements, 2 xenophobic remarks and 1 case of racial stereotyping. Racial slurs have led every category the Commission has published this year, a pattern that holds across the full year-to-date figures too: 77 standalone slurs, 58 prejudicial or insensitive remarks, 45 derogatory statements and 21 xenophobic instances, out of 270 total warnings issued between 1 January and 24 August 2026.
That 270 compares with 194 warnings over the same stretch of 2025, a rise of close to 40 percent. Some of that gap is almost certainly detection catching up with behaviour that was already there, invisible to a monitoring team reading platforms by hand. Some of it may be behaviour getting worse. The Commission's own bulletins do not separate the two, and neither, honestly, can anyone reading from outside the Media Monitoring Unit's own dashboards.
What is not in dispute is the setting this is happening in. Guyana's population, per the 2012 census, the most recent count with a published ethnic breakdown, is 39.83 percent Indo-Guyanese, 29.25 percent Afro-Guyanese, 19.88 percent of mixed heritage and 10.51 percent Indigenous. Guyanese politics has tracked those lines for decades, and Georgetown, the capital where the ERC is headquartered, also hosts the CARICOM Secretariat, giving the Commission's numbers a regional audience well beyond Guyana's own borders.
Inside eMonitor+: The AI Tool the ERC Already Runs
UNDP and the ERC signed a memorandum of understanding on 8 October 2024, ahead of Guyana's 1 September 2025 general election, to deploy eMonitor+ across the country's main social platforms. Gerardo Noto, UNDP's Resident Representative in Guyana, framed the deal at the time as proactive: "by equipping the ERC with the tools and methodologies to combat hate speech and misinformation, we are taking proactive measures to strengthen democracy." ERC Chairman Shaikh Moeenul Hack put the practical case for it more simply: "The E-Monitoring Tool will be instrumental in enhancing our capacity to capture a wider reach of content."
eMonitor+ is not a Guyana-specific build. UNDP has deployed versions of it for election commissions and human rights bodies elsewhere, including in Bolivia, Chile, Ecuador, Panama, Peru and Venezuela, monitoring for hate speech, disinformation, political polarisation and online violence against women, journalists and marginalised groups. Guyana appears to be the only CARICOM member state confirmed to be running it, which makes the Commission's monthly category breakdowns, slurs against prejudice against xenophobia, a level of structured reporting on online ethnic conduct that no other English-speaking Caribbean country currently publishes.
The mechanics are straightforward in outline even if the underlying models are not. Rather than a handful of staff manually reading a sample of posts, the system continuously scans public content across the four platforms, applies a trained classifier to sort matches into the categories the ERC now reports monthly, and surfaces the results for a human reviewer to confirm before a cautionary warning goes out. That last step matters: the AI narrows the field, but a person still decides whether a flagged post crosses the Commission's threshold, which is the difference between an automated content filter and what the ERC is actually running.
The Core Argument
An AI system that reads every public post on four platforms will always outpace a Commission built to investigate each one individually. Guyana's rising warning count is not proof the AI is failing. It is proof the human side of the pipeline was never built for this volume, and closing that gap now matters more than tuning the model further.
Where Detection Outpaces Resolution
The Media Monitoring Unit's warnings are the fast, largely automated track. The ERC's Investigative Unit is the slow one, and the numbers there tell a different story than the monthly bulletins do. As of the Commission's most recent reporting, the Investigative Unit was carrying 87 logged matters: 3 from 2022 and 2023 combined, 14 from 2024, 49 from 2025 and 21 opened so far in 2026. Of those 87, 51 have been closed, 7 completed and 3 withdrawn, which leaves roughly 26 still open, several of them three years old or more.
A cautionary warning from the Media Monitoring Unit is quick: the AI flags a post, a reviewer confirms it, a warning goes to the account holder, often within the same reporting cycle. A case with the Investigative Unit is a different process entirely, involving mediation, conciliation, and in serious matters a referral to the police or the Director of Public Prosecutions, none of which an AI classifier can do on its own. The gap between those two tracks is where Guyana's AI investment is currently paying off least. eMonitor+ makes the detection side faster every month. Nothing comparable has been added to the investigative side, and the 2022-vintage cases still sitting open are the visible result.
Why This Is Not Yet a Solved Problem
Guyana has direct experience of what unaddressed ethnic tension can become. The 1962 to 1964 disturbances in what was then British Guiana, fuelled by rhetoric and rumour between Indo- and Afro-Guyanese communities, left a mark still referenced in the country's politics six decades later. As recently as 2020, a five-month dispute over the general election vote count triggered protests and sharpened ethnic rhetoric before international pressure forced a recount and resolved the standoff. The stakes behind the ERC's monthly numbers are not abstract.
The 1 September 2025 election that prompted the UNDP-ERC partnership returned President Irfaan Ali for a second term, with the PPP/C winning 36 parliamentary seats on roughly 242,000 votes. It also saw a new party, WIN, led by Azruddin Mohamed, draw support across the country's traditional ethnic-voting blocs for the first time in a generation, a development that made the two established parties, and outside observers, more attentive to unity messaging than in past cycles. Poet Carlene Gill-Kerr, quoted by the Carter Center's election mission, described Guyana bluntly as "multiethnic, and often ethnically divided," and said she wanted a result "that would serve everyone well, not just those who look like her."
That same legal environment carries a real risk alongside the opportunity. Guyana's Cybercrime Act 2018 includes a sedition provision, Section 18, that criminalises online content bringing the government into hatred or contempt, and civil society groups raised concern in 2024 that the law could be turned against ordinary political criticism. The ERC's own mandate under Article 212D of the Constitution is narrower and specifically about ethnic relations, and its warnings are an educational measure rather than a criminal charge, a structural distinction that limits but does not remove the risk of an AI classifier misreading sharp political criticism as ethnic hostility if it is not scoped and audited carefully.
What Should Happen Next
Four steps would move Guyana's system from a monitoring programme that works to a full pipeline that resolves what it finds.
First, the ERC and UNDP should extend AI assistance past detection and into triage. A system that can already classify a post as a slur versus a xenophobic remark can also draft case summaries, flag repeat offenders across the Investigative Unit's 87 open and closed matters, and surface which of the roughly 26 still-active cases carry the highest risk of escalating, freeing the Commission's small investigative staff to work the cases that need human judgement most urgently rather than the ones that happen to be next in a queue.
Second, any classifier trained on Guyanese content needs continuous tuning for Guyanese English Creole and the coded, dialect-heavy phrasing that carries ethnic meaning without using an obvious slur, the register most likely to slip past a model built primarily on Spanish-language training data from eMonitor+'s other deployments across Latin America.
Third, CARICOM institutions should treat Guyana's partnership as a template rather than a one-off. Trinidad and Tobago, Suriname and Belize all carry comparable multi-ethnic political dynamics without an equivalent AI monitoring body in place, and the cheapest time to build one is before an election cycle raises the stakes, not after.
Fourth, the ERC should publish a clear, public boundary between what the Media Monitoring Unit flags automatically and what counts as protected political speech, addressing the misuse concern directly rather than leaving it to be inferred from the Cybercrime Act's separate, broader sedition language.
None of these four steps requires new legislation or a new agency. The AI system, the legal mandate and the investigative staff already exist. What is missing is the connective layer between detection and resolution that would let one actually keep pace with the other.
The Question the Numbers Do Not Answer
A rising warning count can mean a country's online discourse is getting worse, or it can mean a country is finally able to see how bad it already was. Guyana's 2026 figures do not settle which explanation is correct, and the ERC has not claimed otherwise. What the numbers do show clearly is that the AI half of this equation is working roughly as designed, catching content at a scale no monitoring team could manage by hand, while the human half, mediation, investigation, resolution, has not scaled with it. Whether that gap closes before Guyana's next election cycle, or simply grows alongside the warning count, is the part of this story still being written.
Frequently Asked Questions
What is Guyana's Ethnic Relations Commission?
The Ethnic Relations Commission, known as the ERC, is a constitutional body established under Guyana's Constitution (Amendment) (No. 2) Act 2000, which inserted Articles 212A to 212F into the Constitution of the Co-operative Republic of Guyana. Its mandate under Article 212D is to promote ethnic harmony, eliminate ethnic discrimination, and safeguard equal opportunity among the country's ethnic communities. The Commission runs a Media Monitoring Unit that scans social media for hate speech and a separate Investigative Unit that handles formal complaints, and it is headquartered at 66 Peter Rose and Anira Streets in Georgetown.
Does the ERC's hate speech monitoring apply to Guyanese posting from the diaspora?
Yes, in practice. The Media Monitoring Unit tracks content on public platforms such as Facebook, Instagram, X and YouTube regardless of where the poster is physically located, since the harm the Commission is mandated to address is the content's presence inside Guyana's online public sphere, not the author's postcode. A large share of Guyana's population lives abroad, particularly in the United States, Canada and the United Kingdom, and diaspora accounts are a routine source of the commentary the Unit flags.
How can someone report suspected ethnic hate speech in Guyana?
A complaint can be lodged directly with the ERC by phone at 231-6265 or 231-6281, by email to info@erc.org.gy, or in person at the Commission's Georgetown office. A formal complaint is separate from what the Media Monitoring Unit catches on its own: content the Unit flags through its own scanning results in a cautionary warning to the poster, while a complaint submitted by a member of the public can open a case with the Investigative Unit, which has the power to mediate, conciliate or refer a matter to the police or the Director of Public Prosecutions.
How much does an AI hate speech monitoring tool like eMonitor+ cost?
Neither the ERC nor UNDP has published a specific licence fee for Guyana's deployment. Tools in this category are typically priced by data volume, the number of platforms and languages monitored, and whether the analysis extends beyond public posts to comment threads and closed groups where permitted. UNDP has positioned eMonitor+ as a shared digital public good available to partner election and human rights bodies across Latin America and the Caribbean rather than a commercial product sold per country.
What is the difference between the ERC's Media Monitoring Unit and its Investigative Unit?
The Media Monitoring Unit scans public social media on an ongoing basis, aided by eMonitor+, and issues cautionary warnings directly to posters, 48 such warnings between 29 July and 24 August 2026 alone. The Investigative Unit is a separate, slower track that opens a formal case file, usually from a member of the public's complaint, and carries the power to mediate between parties or refer a matter for prosecution. As of the Commission's most recent figures, the Investigative Unit was carrying 87 logged matters dating back to 2022, with roughly 26 still unresolved.
Could AI monitoring of online speech in Guyana be used to suppress legitimate criticism?
That risk is real and already documented, though it attaches more directly to a different law than to the ERC's own warnings. Guyana's Cybercrime Act 2018 contains a sedition provision, Section 18, criminalising online content that brings the government into hatred or contempt, and civil society groups raised concern in 2024 that the Act could be turned against ordinary political criticism rather than genuine incitement. The ERC's constitutional mandate is narrower and specifically about ethnic relations, and its cautionary warnings are an educational measure rather than a criminal charge, which limits but does not eliminate the risk that an automated classifier could mistake sharp criticism of a government official for ethnic hostility if it is not carefully scoped and audited.
Is online hate speech illegal in Guyana?
Guyana does not have a single, standalone hate speech statute in the way some countries do. Ethnically abusive content falls partly under the ERC's constitutional mandate, which allows the Commission to issue cautionary warnings and mediate disputes rather than prosecute, and partly under the Cybercrime Act 2018, which criminalises related conduct such as harassment, intimidation and sedition and carries penalties including imprisonment. Most of what the Media Monitoring Unit catches ends in a warning, while a smaller number of more serious matters can be referred onward for criminal prosecution under separate legislation.
Where is AI-assisted hate speech monitoring headed in the Caribbean?
Guyana is, so far, the only CARICOM member state publicly confirmed to be running an AI system built for this purpose, a tool UNDP says is also active in Bolivia, Chile, Ecuador, Panama, Peru and Venezuela. Trinidad and Tobago, Suriname and Belize carry comparable ethnic-plurality dynamics without an equivalent monitoring body in place. The realistic path forward is regional rather than country by country: CARICOM institutions extending the same UNDP partnership to other member states ahead of their next election cycles, rather than each government building its own version only after a crisis has already made the case for it.