Story perspectives

AI Datasets Raise Consent Concerns in Content Creation

9/11/2025

Summary

Story summary

Runway AI's "Gen-3" model was trained using 3,967 YouTube channels identified for high-quality content, according to The Atlantic's investigation.
Snap's Panda-70M dataset, released in 2024, includes 3.8 million YouTube videos divided into 70.7 million AI-captioned clips.
Kuaishou Technology's Koala-36M dataset, also released in 2024, features over 2.4 million videos and 36 million clips.
Researchers from the Shanghai Academy of AI for Science developed VidGen-1M, a dataset of 1 million clips sourced from Microsoft's HD-VILA-100M.
All datasets are available on Hugging Face, prompting concerns about consent and the use of creators' content for AI training.

History

Historical Context

The European Union implemented the General Data Protection Regulation (GDPR) in 2018 to enhance data privacy and user consent. This regulation has significantly impacted tech companies, including Snap and Kuaishou Technology, which have released extensive datasets. As AI technologies advance, ethical concerns regarding the use of creators' content without consent continue to challenge legal frameworks and the rights of content creators globally.

1 of 2

Story summary

Runway AI's "Gen-3" model was trained using 3,967 YouTube channels identified for high-quality content, according to The Atlantic's investigation.
Snap's Panda-70M dataset, released in 2024, includes 3.8 million YouTube videos divided into 70.7 million AI-captioned clips.
Kuaishou Technology's Koala-36M dataset, also released in 2024, features over 2.4 million videos and 36 million clips.
Researchers from the Shanghai Academy of AI for Science developed VidGen-1M, a dataset of 1 million clips sourced from Microsoft's HD-VILA-100M.
All datasets are available on Hugging Face, prompting concerns about consent and the use of creators' content for AI training.

1 / 2