YouTube used AI to alter creators' videos without consent
YouTube secretly used machine learning to enhance YouTube Shorts videos by unblurring, denoising and clarifying footage without informing or obtaining consent from creators. The undisclosed alterations caused visible artifacts that creators said made their work appear AI-generated and untrustworthy, and YouTube only confirmed the experiment after months of complaints.
What happened
YouTube secretly used artificial intelligence to enhance and alter a limited number of YouTube Shorts videos without informing creators or obtaining consent.
The AI system applied changes including unblurring, denoising, and enhancing clarity, resulting in visible alterations such as sharper details, smoother skin textures, warped ears, and over-sharpened wrinkles in fabric. YouTube confirmed the experiment ran on select YouTube Shorts using traditional machine learning technology similar to smartphone video processing.
Reported impact
Creators reported that the AI alterations made their videos look unnatural and AI-generated, with subtle visual artifacts that appeared worrying for audience trust. Musician Rick Beato noticed his hair looked strange and his skin appeared to have makeup, while fellow YouTuber Rhett Shull complained the over-sharpening gave videos an unwelcome AI-generated feeling and potentially misrepresented his work.
Creators expressed concern about loss of control over their original content and the potential erosion of trust with their audiences. Shull's video documenting the issue accumulated over 500,000 views, with complaints on social media dating back to at least June.
Index analysis
YouTube's experiment raised broader questions about digital rights, the authenticity of online content, and the growing influence of AI in shaping reality. The undisclosed alterations highlighted tensions between platform-driven optimization and creator autonomy.
Governance controls
YouTube acknowledged the experiment only after months of rumors and complaints from creators. The company described the system as traditional machine learning rather than generative AI, and stated it was testing on a limited number of videos on YouTube Shorts.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.