Google Trains AI Models on YouTube Videos Without Full Creator Awareness

Google is leveraging its vast YouTube library—approximately 20 billion videos—to train AI models like Gemini and Veo 3. While the company claims compliance with creator agreements, billions of minutes of content are being repurposed for machine learning. YouTube defends the practice as historically consistent with product improvement, though creators lack opt-out mechanisms for AI training.

Concerns escalate as tools like Vermillio's Trace ID detect striking similarities between AI-generated and original content. A Veo 3 output scored 71% visual and over 90% audio alignment with a video by creator Brodie Moss. The tension reflects broader debates over intellectual property in the AI era, with some creators embracing competition and others alleging unauthorized use.