Brand safety AI systems using keyword-based detection incorrectly classified legitimate news articles as 'brand unsafe', blocking ads from mainstream publishers and redirecting revenue to lower-quality sites.
Brand safety AI systems deployed by vendors like Moat and Comscore were designed to protect advertisers from placing ads on inappropriate content by analyzing web pages and classifying them as 'brand safe' or 'brand unsafe'. However, these systems relied heavily on simple keyword lists rather than advanced AI as claimed in their marketing materials. The systems incorrectly flagged legitimate content on major news sites like New York Times, Wall Street Journal, and Economist as 'brand unsafe' simply due to the presence of words like 'covid-19', 'coronavirus', 'death', or 'weapons'. Research by Adalytics found that 21% of Economist articles, 30% of New York Times and Wall Street Journal articles, and 52% of Vice articles were incorrectly labeled as brand unsafe. When comparing two major vendors, Moat and Comscore disagreed about safety classifications 40% of the time on Wall Street Journal articles. The misclassifications particularly affected serious journalism covering topics like Middle East affairs, obituaries, and political events. As a result of these incorrect classifications, advertisers' ads were blocked from legitimate news sites and their ad spending was redirected to lower-quality sites in programmatic channels, causing significant revenue loss to mainstream publishers.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
AI systems that fail to perform reliably or effectively under varying conditions, exposing them to errors and failures that can have significant consequences, especially in critical applications or areas that require moral reasoning.
AI system
Due to a decision or action made by an AI system
Unintentional
Due to an unexpected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed
No population impact data reported.