Music sector battles AI-generated deepfake songs
Generative AI music tools have been used to create deepfake songs mimicking real artists without consent, prompting the music industry to pursue takedowns and prompting platforms like Spotify and YouTube to develop detection technology, though observers say the problem outpaces existing countermeasures.
What happened
Generative AI music tools have created deepfake songs without artist consent, with Sony Music reporting requests to remove 75,000 fake videos from the internet. The tools generate songs that mimic real artists, including fake rap songs attributed to Tupac about pizza and K-pop songs imitating Ariana Grande that she never performed.
Music industry actors are working through digital platforms, courts, and elected officials to prevent generative AI tools from stealing musical content, but results remain unsatisfactory overall.
Reported impact
The widespread creation of fake musical content represents a scale of the phenomenon that concerns the music sector. Unauthorized use of artist voices and styles threatens musicians' control over their intellectual property and public reputation.
Index analysis
Technology companies specializing in audio detection, such as Bandrop, confirm that AI-generated songs contain subtle errors in frequency, rhythm, and digital signature not found in human-created works, making detection technically possible. Spotify's regulatory policy head stated the company takes the matter seriously and is developing new tools to improve detection of AI-generated fake content.
YouTube confirmed it is improving its detection technology with partners and may announce updates in the coming weeks, though analyst Jeremy Goldman noted that bad actors advance one step ahead, with platforms like YouTube under pressure to solve the problem to prevent their services from becoming an AI nightmare.
Governance controls
The music industry is pursuing takedown requests, with Sony Music requesting removal of 75,000 fake videos. YouTube and Spotify have stated they are developing or improving detection tools and working with partners on the issue.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.