The Internet is awash in AI content, and it’s not always easy to tell it apart from genuine human creations. While images and videos are perhaps the most obvious type of AI slop, streaming music services like Spotify are also being inundated with AI-generated tunes. Suno is one of the most prolific sources of this AI music, but the company now says it’s looking to clean up its act with watermarks and policy changes.
In a blog post announcing the plans, Suno CEO and co-founder Mikey Shulman says this is a necessary change to meet “emerging industry standards” for the labeling of AI content. Soon, the company will add watermarks to all audio outputs from its models, giving platforms the option to mark content as AI or block it entirely. Shulman doesn’t say if Suno will rely on an in-house solution or a ready-made solution like Google’s SynthID.
Google recently began licensing SynthID to other companies. Google says the tech has been used to label 60,000 years’ worth of audio generated by its Gemini models, plus more than 100 billion images and videos.
Whatever technology Suno uses, the idea is that every track it generates will have an invisible code embedded in the waveform. A detector trained on that signature will be able to identify it, allowing Suno’s partners to combat “fraud and misuse.” Shulman notes the watermarking tech Suno intends to use is “durable and resistant to tampering” without affecting quality.
However, no system is perfect. If someone does manage to break Suno’s watermarks, all the tracks produced to that point could be stripped of their AI labels. And this doesn’t solve the problem of proliferating AI music. There are other platforms that don’t have plans for watermarking, and open models could allow people to generate unlabeled AI music on their own hardware. That’s a problem with all efforts to watermark AI content.