Bellingcat Uses AI to Streamline Detection of Civilian Harm in Ukraine Conflict

Bellingcat, a renowned open-source intelligence organization, has developed and tested an innovative machine learning model designed to identify incidents of civilian harm from social media posts. Between February 2022 and September 2025, the organization's staff and volunteers collected over 2,500 verified instances of such harm in Ukraine, marking a significant step forward in documenting the impact of conflict on civilians. The new AI model analyzes Telegram posts to predict those likely to contain evidence of civilian harm, significantly reducing the time required for manual searches.

The methodology involves creating a dataset that includes both confirmed cases of civilian harm and a larger number of non-harmful posts to train the machine learning algorithm effectively. This approach ensures that the model can accurately distinguish between relevant and irrelevant content. By enriching each post with metadata such as publication time, reactions, and textual context, researchers aim to provide the AI with comprehensive information necessary for accurate predictions.

This development is particularly timely given ongoing conflicts in regions like Sudan and parts of the Middle East, where civilian harm remains a critical issue. The use of machine learning not only expedites the identification process but also frees up human resources to focus on verifying these incidents rather than searching through vast amounts of data. Bellingcat's work underscores the potential for AI to play a crucial role in humanitarian efforts and conflict documentation, offering hope that similar methodologies can be applied globally to protect civilians caught in war zones.