AnalysisAI ModelsAugust 18, 2026

AI Observatory finds AI usage reports are skewed to work, missing half of conversations

Read original source →technologyreview.com

The Stanford-led AI Observatory analyzed consent-based conversations with models like Claude and Gemini, finding 48% of conversations would be filtered out by Anthropic's work-focused methods, and that health, relationships, and harassment topics appear far more often than company reports suggest.

People · Anka Reuel

1 source

More stories today

Open the live feed