AddressHate detects hate speech, particularly its coded and disguised forms that existing tools miss. Our suite of tools enable institutions, policymakers, and digital platforms to stop digital hate from turning into real-world harm.
A world where hateful online discourse is stopped before it becomes real-world harm.
AddressHate is an independent nonprofit building technology to serve the public good. We are developing AI to detect harmful language online that current moderation tools miss.
AddressHate’s AI model detects hate more efficiently and accurately than existing models because it understands hate at the narrative level. We are creating superior moderation tools by translating leading academic institutions’ independent, peer-reviewed research into expert-labeled data to train detection models.
Rigor: Every finding is grounded in peer-reviewed research and methodological transparency.
Independence: Editorial and institutional independence from government, advocacy groups, political actors, and platforms.
Urgency: Digital hate causes real-world harm that demands immediate action.
Collaboration: The problem is too complex for any one discipline to solve alone.
Precision: We go deeper than keywords and slurs to catch what others miss.
Real online content and comments are labeled by experts against a comprehensive set of categories, which were developed by leading hate scholars and linguists to identify specific stereotypes that express antisemitic and other hate ideologies. This data is used to train our AddressHate AI model, train existing models (Claude, OpenAI) to improve their accuracy in detecting hate and highlight their current blind spots, and provide researchers in the field insights missed by current tools.
Technology designed to detect which narrative is present in online discourse. It is built using the library of expert-labeled training data (see above) and built by our machine learning scientists. This AI model can be used as a resource by partner organizations to detect implicit hate on social media platforms and AI model outputs for the purposes of furthering research in the space.
An independent measure of how social media and AI companies monitor hate and enforce their own safety policies.
Using our AddressHate AI model and library of expert-labeled training data, we surface how dangerous stereotypes take root and spread, so newsrooms, policymakers, and educators can act on it.
Annotating data means having subject-matter experts read real online content and flag hateful narratives so that an AI model can learn to recognize it. It is one of the most time- and cost-intensive parts of building an effective AI model for hate detection. We developed a tool that simplifies the labeling process for nonprofits, companies, and researchers tackling online hate.
The first academic journal devoted to digital hate, built to bring historians, linguists, sociologists, computer scientists, and policymakers into one field to address this interdisciplinary problem.