AI music generator Suno has a major problem: its copyright protection system is basically useless. The platform, which promises to block users from uploading copyrighted material, turns out to be trivially easy to fool with minimal effort and free software. The result? AI-generated knockoffs of hits from Beyoncé, Black Sabbath, and Aqua that sound alarmingly close to the originals, raising serious questions about AI’s ability to self-police copyright infringement at a time when the music industry is already suing the company.

Suno, the AI music generation platform that lets users create songs from text prompts, has positioned itself as a responsible player in the generative AI space. The company’s policy explicitly states it doesn’t permit copyrighted material. Users can upload their own tracks to remix or set original lyrics to AI-generated music, but the system is supposed to recognize and block other people’s songs and lyrics.

Turns out that’s not quite how it works in practice. According to The Verge’s investigation, Suno’s copyright filters are incredibly easy to fool, and with minimal effort and some free software, the platform will happily spit out AI-generated imitations of popular songs that sound alarmingly close to the originals.

The publication tested the system with iconic tracks like Beyoncé’s “Freedom,” Black Sabbath’s “Paranoid,” and Aqua’s “Barbie Girl.” While most people can likely tell the difference between the AI versions and the real thing, the resemblance is close enough to be deeply concerning for artists and rights holders. The AI-generated covers capture distinctive vocal styles, melodic structures, and production elements that make these songs instantly recognizable.

This isn’t just an academic exercise. Suno is already fighting a high-stakes legal battle with the music industry. The Recording Industry Association of America sued Suno alongside competitor Udio in 2024, alleging the companies trained their AI models on copyrighted music without permission. The lawsuit claims both platforms engaged in “stream ripping” from YouTube and other sources to build their training datasets.