Suno is a copyright disaster for music.

The policy of AI music platform Suno states that it does not allow the use of copyrighted content. Users can upload their own songs for remixing or pair original lyrics with AI-produced melodies. However, the system is designed to identify and prevent the use of others’ music or lyrics. While perfection is unattainable, it appears that Suno’s copyright filters can be easily circumvented.

With minimal input and some free software, Suno can generate AI-driven reproductions of popular tracks like Beyoncé‘s “Freedom,” Black Sabbath’s “Paranoid,” and Aqua’s “Barbie Girl” that closely resemble the originals. While the majority may distinguish them, a few could be mistaken for unofficial versions or alternative takes upon casual listening. Furthermore, it is conceivable that someone could monetize these uncanny replicas by exporting and posting them to streaming platforms. Suno opted not to provide comment for this report.

Creating these renditions involves utilizing Suno Studio, which is included in the company’s $24-a-month Premier Plan. Instead of generating an entire song through text prompts, Suno Studio allows uploads of tracks for modification or covering. It’s likely to reject a well-known song without alterations. However, employing a basic free tool such as Audacity to slow down or speed up a track can frequently get around the system, and adding a burst of white noise to the beginning and end seems to ensure success. Once in Suno Studio, one can restore the original speed and remove the white noise, allowing the copyrighted song to serve as a foundation for new AI-generated music.

If you produce a cover of the uploaded audio without altering styles, Suno essentially outputs the original instrumentals with minimal adjustments to the sound palette, particularly with model 4.5 or 4.5+. Model v5 is slightly more liberal with creative liberties, incorporating chugging guitar sounds and galloping piano in “Freedom” and transforming the Dead Kennedys’ “California Über Alles” into a jig led by violins.

Suno allows you to incorporate vocals by either generating lyrics or inputting text, and again, it’s meant to block copyrighted material. If you copy and paste the official lyrics from Genius into the system, Suno will flag them and produce nonsensical vocals. Nevertheless, slightly modifying lyrics can successfully bypass this restriction as well.

I was successful in deceiving Suno Studio by altering the spelling of a few words in “Freedom”—transforming “rain on this bitter love” into “reign on” and “tell the sweet I’m new” into “tell the suite”—and aside from the first verse and chorus, I didn’t need to apply further changes. The voice closely replicates the original, drawing forth slightly altered versions of Ozzy or Beyoncé.

Indie musicians may not enjoy that level of protection. One of my own songs was able to pass through the copyright filter while I was experimenting with v5 of the model. I also successfully got tracks by singer-songwriter Matt Wilson, Charles Bissell’s “Car Colors,” and experimental artist Claire Rousay past Suno’s copyright detection setup without making any adjustments whatsoever. Independent artists on smaller labels or those distributing via Bandcamp or services like DistroKid are particularly likely to fall through the cracks; DistroKid and CD Baby did not respond to requests for comment.

The output of these AI-produced covers resides firmly within the uncanny valley. The songs they are imitating are unmistakable: the “Paranoid” riff is clear, and “Freedom” is distinctly recognizable from the outset when the marching snare hits commence. Yet, there’s a certain lifelessness present. Although AI Ozzy sounds incredibly accurate, it lacks depth and dynamics, resulting in an impression of being a mere imitation of a human, not an authentic performance.

The instrumentals likewise eliminate any intriguing artistic decisions made by the originals or replicate them in unexciting ways. A non-jig version of “California Über Alles” has most of its rough edges smoothed out, giving it the sound of a wedding band rendition. Pink Floyd’s “Another Brick in the Wall” transforms from an exploration of doom disco to just bland dancefloor filler. While it captures David Gilmour’s guitar tone, it omits any sense of phrasing or evolution, turning the solo into a mindless stream of notes.

Creating unauthorized covers contravenes both the intended function of Suno and the terms of service. Additionally, it appears that Suno only scans tracks upon upload; it seems that outputs are not rechecked for potential violations or rescanned prior to exporting. From here, monetizing Suno-created covers is a straightforward process. Those who would exploit AI-generated content could upload it through a distribution service like DistroKid and benefit financially from other creators’ music without compensating the original composers as would typically happen with a cover. Independent artists appear particularly at risk.

Folk artist Murphy Campbell recently encountered this when someone uploaded what appeared to be AI-created covers of her songs from YouTube to her Spotify account. (It’s unclear which system they were produced by.) Shortly after, distributor Vydia placed copyright claims against her YouTube content and began collecting royalties. To illustrate how flawed the entire system is, the songs Vydia successfully claimed copyright on are all public domain. Spotify eventually removed the AI covers, and Vydia has retracted its copyright claims, but this only occurred following a social media campaign initiated by Campbell. Vydia claims that the two situations are unrelated and asserts no connection to the AI covers of Campbell’s creations.

AI imitations pose a threat to other artists as well. Experimental composer William Basinski and indie rock band King Gizzard and The Lizard Wizard have seen replicas slip through various filters and reach streaming platforms like Spotify. On occasion, these counterfeit tracks can drain views directly from the artist’s own page. In an environment where payouts are already scarily minimal — Spotify mandates a minimum of 1,000 streams to receive payment — lesser-known musicians are particularly affected.

Platforms such as Deezer, Qobuz, and Spotify have implemented measures to address spammy AI and imposters. Spotify representative Chris Macowski informed The Verge that the company “takes artist rights seriously and tackles the issue from multiple perspectives. This includes protections to prevent unauthorized content from being uploaded in the initial place, as well as systems capable of identifying duplicate or highly similar tracks. These mechanisms are supplemented by human reviews to ensure accuracy.” Nonetheless, no system is infallible, and managing an influx of AI-driven content made possible by platforms like Suno presents a significant challenge.

Macowski acknowledged the technical challenges faced, stating, “It’s an area we are still investing in and evolving, especially as new technologies surface.”

Suno constitutes only one part of an evidently malfunctioning system. However, it remains an avenue where artists possess particularly restricted options for addressing grievances. Bands can reach out to Spotify to have AI forgeries removed from their profiles. It’s more challenging to determine how these forgeries are created and whether they result from failures in Suno’s regulatory processes. As of now, Suno’s reply is silence.