As streaming catalogs are inundated with a growing volume of AI music, platforms face the mounting challenge of protecting the customer experience. As major players like Deezer launch their own AI detection models, the focus for many in the industry is shifting from merely identifying AI to mitigating its potential damage to listener trust.
Rasha Rahman is a frontend-focused software engineer and indie hacker working at the center of this issue. He's the founder of HumanStandard, a platform designed to detect AI-generated music. He says while there's plenty of discourse around protecting creators' authorship, there's not enough discussion about AI's impact on the listener experience.
"The listener is going to have a different experience with your catalog because of the increase in fully generated AI music," he says. "It's really important that the experience of that catalog stays the same or doesn't decrease in quality." Rahman breaks the problem into two core challenges. The first, he explains, is the ethical dilemma around provenance as the origins of new music become increasingly opaque. The second is the more subtle but equally damaging business risk of eroding listener trust and a degraded user experience.
- The copyright question: The use of copyrighted music or voice recordings to train AI models is currently in a legal gray area, which Rahman says could present significant issues down the road. "A lot of the current audio-generating models were trained on copyrighted works, creating a period of music that is unregulated. A consumer would prefer not to listen to music that's stolen. That's the first ethical concern."
- Listener liability: The second concern, he explains, is what happens to subscription platform listeners when they're suddenly exposed to a heavy amount of content made partially or completely with AI. "When I talk to product managers and founders of these companies, they tell me they don't want their users to leave because there's more AI content in their feeds."
Rahman says whether or not listeners can tell the difference between human-made and AI-made content, origin matters to them. He proved this idea with a viral TikTok and Instagram series dubbed Human Versus AI. "I show people a human song and an AI song, and I ask them which one is AI. This got millions of views." He says the series helped him build a network of more than 40,000 musicians, creators, and fans, all of whom care deeply about the issue of provenance.
While many in the industry point to solutions like transparency and labeling, Rahman feels these measures don't go far enough. In his view, they fail to address a more fundamental problem with AI’s impact on creative credit. "Human authorship has been destroyed, and people don't want to admit it. When AI generates something, it never tells you it came from a specific human. That person is just gone. In contrast, when you hold a vinyl, you know exactly whose art it is."
- The open source problem: This viewpoint has immediate practical consequences. Rahman points out that the rise of powerful, untraceable open-source models can allow large amounts of unattributed AI music to slip through the cracks. "Anyone can download a bunch of songs, train a model, and generate music with no attribution connected to it." This, he says, makes policy-based solutions difficult to enforce. "I believe the only solution is the black box, high-defense method. It's a cat and mouse situation now."
- Thought vs. prompt: Rahman acknowledges that the line between human and machine creativity exists on a spectrum. He defines the fundamental difference by where the process begins. "The danger is externalizing your creativity before you've internalized it. The experience of playing keys on a keyboard, of thinking and creating a sound from within, is fundamentally different than just writing a prompt. Creative agency is what makes us human, and we can't let that be automated away."
That philosophy is the driving force behind his work. Rahman is piloting HumanStandard with music companies with the goal of scanning up to a million songs per day. The platform's detection model is a real-world application of his principles. "Our model is trained on human work, that the creators have given full consent to use, and in return will get a revenue share."
Ultimately, Rahman approaches his work in human-centric terms, noting that it's easy to lose perspective when you're too close to the technology. "A lot of people are in the matrix when it comes to AI," he says. "They're so tapped into the AI world that they've forgotten that there are humans that still exist. I think music and the community that can come from it is so beautiful, and saving and preserving human art is really, really important."