Armenia’s New Hate Speech Detector

“We are seeing more hate, more coordinated attacks and more people, from journalists to activists to ordinary users, being pushed out of online conversations because of it,” said Alexandr Martirosyan, project lead at All Equal Rights Foundation (AERF), the local human rights and digital security NGO that developed the moderation tool.

The ModerAItor platform recognizes Armenian-language content written in Armenian, Latin transliteration and Cyrillic transliteration, addressing a gap not covered by existing moderation systems.

It was developed after AERF commissioned research that found women in public and political life were particularly vulnerable to online abuse, including profanity, sexual harassment and calls for violence. It also found that online hate speech discouraged women’s participation in public life and contributed to the normalization of discriminatory attitudes.

“ModerAltor fills a critical void that Big Tech consistently ignores, because effective hate-speech filters directly threaten the outrage-driven engagement that fuels their business models.” said Artur Papyan, the partner lead at the Media Diversity Institute, an international organization working on accurate, ethical reporting which collaborated on the project.

Designed as a decision-support tool rather than an automated enforcement mechanism, moderAItor helps users identify potentially harmful comments while leaving moderation decisions to human reviewers.

“What is different about moderAltor is the context it was built for… the way people mix Armenian, Latin and Cyrillic scripts in real conversations,” explained technical lead Arman Zohrabyan. “The categories the AI uses come from an editorial moderation policy we developed ourselves, grounded in Armenian law and international standards.”

Developed with support from IWPR’s Building Resilience in the Eastern Neighbourhood (BREN II) program, moderAItor was piloted by media organizations analyzing more than 150,000 comments and moderating around 43,000 of them—approximately 28.7 per cent of the content reviewed.

This demonstrated the tool’s potential to support organizations managing high volumes of online engagement, particularly during periods when hate speech and online harassment intensify.

A major milestone came in February 2026, when the Media Ethics Observatory approved the project’s comment moderation policy and incorporated it into Armenia’s media self-regulation framework, making it available as professional guidance for around 100 organizations.