Creators caught off guard as Google trains AI on YouTube

Google is feeding its massive vault of YouTube videos directly into its AI training pipeline, powering both Gemini and its flagship video generator Veo 3.

According to a report from CNBC, Google confirmed it uses YouTube content for AI training, though it claims it only taps a subset of videos and honors existing partner agreements. An unauthorized insider noted the company is leaning on its catalog of 20 billion videos to get the job done.

YouTube didn't disclose exact numbers, but experts say training on just 1% of that catalog equals 2.3 billion minutes of footage. That is roughly 40 times the training data utilized by competing AI video models.

Block Apple, but feed Google

Creators upload roughly 20 million videos to YouTube every single day. While YouTube allows creators to opt out of third-party AI training from companies like Apple, Amazon, and Nvidia, there is no way to opt out of feeding Google itself.

Technically, Google covered its legal bases long ago. When users upload content, they agree to terms granting YouTube a worldwide, royalty-free, transferable license. YouTube also noted in a quiet blog post last September that content could improve services via machine learning.

Yet several top creators and IP attorneys told CNBC they were completely unaware their work was being repurposed to build Veo 3. Identity protection firm Vermillio even ran Veo 3 outputs through its detection tool and found video closely matching creator Brodie Moss, registering an audio similarity score over 90 out of 100.

Reactions remain split. Megacreator Sam Beres, who has 10 million subscribers, called it an exciting inevitable and treats it as friendly competition. Meanwhile, digital identity firms argue creators are effectively forced to train cheap synthetic clones of themselves without compensation.

Why it matters

This marks a fundamental shift in the social contract between Big Tech platforms and the creators who built them.

For years, creators traded their labor for platform reach and ad revenue. Now, Google has converted that two-decade creative backlog into an unmatched moat for multimodal AI. While rivals like OpenAI and Runway scrape the public web or scramble for studio licensing deals, Google is sitting on 20 billion videos with virtually no data acquisition costs.

For creators and video editors, opting out simply is not an option if you want to remain relevant on the world's biggest video site. You are essentially paying rent to a landlord who uses your interior design to build prefab houses next door. We think this lopsided dynamic will inevitably accelerate regulatory pushback and copyright litigation, especially with Hollywood giants like Disney and Universal already taking Midjourney to court.