Something interesting is happening in the world of AI music. What was once a Wild West of devil-may-care capitalism has recently taken on a decidedly conciliatory tone. Ethics, responsibility, transparency, principles – these are the buzzwords peppered through the press releases and mission statements of major players and start-ups alike.
It would seem that the AI music industry is turning over a new leaf and offering musicians an olive branch, and no company encapsulates this trend more than the music generation platform Suno.
Long the poster child for bad behaviour among skeptics of AI’s role in music-making, Suno have previously admitted to training their early models on vast amounts of music scraped from the internet without permission, much of it protected by copyright.
Latest Videos FromMusicRadar
However, CEO Mikey Shulman recently shared a blog post outlining a newly responsible approach. (It’s a remarkable pivot from Shulman, whose public persona has formerly consisted of god-tier trolling.) Moreover, the company is cleaning up its act with the introduction of new industry-backed models trained on licensed data, download limits and a watermarking system designed to help streaming platforms identify AI-generated songs.
“The future of music needs to be built in partnership with the artists, industry and music ecosystem that made music what it is today,” Suno’s Chief Product Officer Jack Brody told MusicRadar ahead of the V6 launch.
“These models were developed in partnership with the industry and informed by feedback from our community of users and artists and musicians at every level. I think this really is a huge moment for Suno, and it marks the beginning of a new chapter.”
What could have prompted this industry-wide about-face? The cumulative weight of lawsuits, labels and publishers suddenly looking to collaborate rather than litigate, growing stocks of ‘in-house’ data created and owned by the companies themselves – all of this has made the practice of scraping copyrighted music from the open internet less attractive.
But perhaps the most important factor is the society-wide AI backlash that is now beginning to bubble over. Even feet-draggers like Spotify are finally combating the AI slop that is flooding their platform.
The public sentiment is overwhelming. A 2025 study by BPI found that 82% of respondents believed that human creativity is essential to music, 80% viewed human-made music as more valuable than AI-generated music, and that 81% wanted fully AI-generated music to be clearly labelled. At the same time, Muse Group (the company behind Ultimate Guitar, MuseScore, and Audacity) surveyed 1200 musicians using their services and found that while 78% were open to using some form of AI, 81% agreed that the industry needed stronger rules and transparency.
So if consumers and musicians want guilt-free AI, the real question is: which companies can they actually trust, and which ones are merely paying lip service?
The unfortunate answer is that there is no easy way to tell. Talk is cheap, feel-good language is on-trend, and consumers hoping to vote with their wallets need to look past vague statements and seek out companies that provide detailed and concrete information regarding their AI policies.
Since its founding in 2023, Voice Swap AI has been one of the most transparent music companies working with AI, taking what CEO Ausrine Skarnulyte describes as an “accountable” approach to development.
Working directly with artists, the platform offers authorised voice models that can be used to create demos and legally licensed for music releases. Their datasets have been certified by the third-party organisation Fairly Trained, outputs are embedded with a ‘high-frequency watermark’ that allows any misuse to be traced back to the offending user, and 50% of the platform’s subscription revenue and 80% of licensing revenue goes directly to artists who have contributed to the platform.
“We chose a rights-first business, even though it was a little bit more painful and slower to establish, because most of the team comes from very deep-rooted music industry experience,” Skarnulyte says of the founding ethos behind the company.
“Building technology is hard but establishing rights around it is even harder,” Skarnulyte continues. “So before we even built the service, first we identified what processes we needed to put in place, what contractual arrangements needed to be made with the singers, what was the usage, what were the licensing terms, what were revocation terms? First, we agreed on those, then we started building the model and the infrastructure around it.”
Voice Swap AI is something of a case-study in what responsible AI development can look like; they also serve as a stark reminder of how few specifics most other businesses choose to share about their practices.
“’Ethical AI’ risks becoming a marketing term now,” warns Skarnulyte. “At the end of the day, ethics isn’t a positioning statement. It is a set of operational decisions you should be able to explain – and few companies actually do.”
The problem, says Skarnulyte, is not individual bad actors per se, but an almost structural lack of clarity around what an ethical approach even is. “If every company can define ethics for itself then businesses with fundamentally different training, content, or deployment practices can make the same claim. At that point, the label itself stops being informative.”
For Geraldo Ramos, CEO and co-founder at Moises AI, it’s not enough to focus only on how training data is obtained – it also matters what kind of products you are making.
“The goal is always to be a tool for musicians and never to go for the whole creation of a song”
“We want to be inserted in the workflow of the musicians as opposed to being something that can bypass that,” states Ramos. “On the data, we decided to train with things that we license or that we create. On the product side, we want to be complementary to a human producer, to make sure that the music qualifies as human-created in the end.”
Moises AI has long been a leader in stem separation, but also offers users a suite of AI-powered creative tools. More recently they’ve introduced audio generation for stems, but maintain that they will always draw a line at fully-fledged song generation.
“The goal is always to be a tool for musicians and never to go for the whole creation of a song,” emphasises Ramos. “We have that constraint, and we will keep it, because we think the way that we do generative is no different than loops.”
Ramos goes so far as to question whether full-on song generators can ever be justified: “Even if you license all the data, one can argue that creating a model which generates the full song is unethical.”
When asked to respond to that viewpoint and the broader AI backlash, Brody acknowledges the shifts in consumer sentiment but maintains that more music made by more people is ultimately a positive, reminding us that there is clear demand for what Suno offers.
“I think we have a real responsibility to hear this feedback,” allows Brody. “To hear the criticisms, to hear the concerns, and to hear from the artists and the creatives who might feel threatened by this technology. It’s an important conversation, and we certainly are not dismissing it here.
“At the same time, the majority of our team are artists and musicians who have been making music their entire lives. The people we speak with every day when we’re building these products are musicians and artists whose entire livelihood is built around creating music. The core of what we hope to build is a tool to empower creativity, and our belief is that it is a good thing if more people are able to experience the joy of making music.”
“There will always be a vocal minority with any new technology,” Brody continues. “We’ve seen it time and time again in music, probably more so than in other mediums, and the pattern usually plays out the same way. Two years ago, there weren’t a lot of artists and musicians embracing this technology. Now there aren’t that many studio sessions in the world that are happening without the usage of some of these tools, the majority of which end up being Suno.”
Beyond defining what ethical AI for music looks like, there’s also the more pragmatic question of how to encourage the industry to adopt such practices. On both counts, few figures have been as influential as Paul McCabe, Senior Vice President of Research and Innovation at the Roland Future Design Lab. With 40 years working in music technology and almost 35 at Roland, he’s seen plenty of new technologies hit the market, but when OpenAI first dropped ChatGPT it provoked a “period of existential crisis” for McCabe.
The sprawling implications of generative AI eventually led McCabe to create the Principles for Music Creation with AI in 2022, first as an internal effort to establish a governance framework for Roland, and later as an industry-wide attempt to build consensus around AI development.
“I was actually thinking back to the early 1980s when the musical instruments industry came together to develop and adopt MIDI,” McCabe says of his early thinking on the issue. “That still stands as an example of industrial unity without equal. Could we do this again?”
Among other things, the initiative calls for respecting copyright protections and data transparency, centering humanity at the core of music-making. It’s now backed by over 150 major brands, labels and institutions, and McCabe hopes to see that number climb to 500 supporters. Of course, in contrast with the adoption of MIDI, there are major economic incentives to take shortcuts when it comes to AI. McCabe is clear-eyed about the limits of trusting corporations to do the right thing, and acknowledges that internal governance must be combined with enforcement mechanisms.
Skarnulyte believes that if enough stakeholders can come together to agree on definitions and enforcements, then we could see the AI industry adopt a system reminiscent of B Corp Certification, which verifies a business’s social and environmental practices, or the EU’s protected designation of origin laws, which safeguard certain foods against imitation.
“It would be much easier for AI companies that are labelled ethical to be marketed,” she notes. “And for the user to make a more informed decision on what tools and services to use.”
Actually passing laws with teeth is where things get tricky. “Legislation is always a couple of years behind,” says Ron Gubitz, Executive Director of the Music Artists Coalition. “Will the government be able to keep up? I think that’s difficult.”
“If I’m a Suno or a Udio right now, and every day I’m catching a haymaker from a different angle, that’s a real [business] incentive”
Describing the organisation’s role as the “pugilistic representative for artists”, Gubitz calls for “clarity, consent, and compensation” as the basis for any definition of ethical AI. While he does point to some bright spots, such as the TRAIN Act – a proposed piece of U.S. legislation that would, among other things, allow training data to be subpoenaed by rights holders alleging copyright infringement – he believes the current shifts are being driven by laws that are already on the books.
“Litigation is expensive and annoying and it seems unending,” Gubitz points out. “If I’m a Suno or a Udio right now, and every day I’m catching a haymaker from a different angle, that’s a real [business] incentive.”
Suno’s new V6 models are at least in part a result of a lawsuit initially brought by the world’s three biggest labels: Sony Music, Universal Music Group and Warner Music Group. When WMG unexpectedly opted to settle their case and partner with Suno – along with BMG, the world’s fourth-largest music company, and Believe, a major global distributor – it paved the way for these new officially licensed models.
“The V6 family was trained from scratch,” states Brody. “It is not trained on the same data as the models that preceded it. It’s trained on a combination of data we’ve licensed with these partnerships and user data.”
Buy-in from labels is a step in the right direction, however, we have no insight into the artists or songs that were included in the training data of the new models, and it’s not clear whether any of the artists involved were remunerated for the use of their work. The American Federation of Musicians is now suing WMG and Universal Music Group over the licensing deals cut with Suno and Udio respectively.
It’s also important to note that while Suno emphasises the start of a new chapter, the company is not conceding that it did anything wrong in the first place. “We believe the way that we’ve trained these models is fair use,” states Brody. “At the same time, we have always had the intent to build in partnership with the industry. We have never had the mindset of ‘we are here to disrupt, and replace, and own the future.’”
“There’s a big difference between generating a song and being able to get separated drums”
Lost in all this is the simple fact that the musical applications of AI are incredibly diverse, and many of them are extremely useful. “It is a bit frustrating,” admits Ramos. “Because some people don’t even know that stem separation is AI. When you say music and AI, it’s always connected to full song generation – but there’s a big difference between generating a song and being able to get separated drums.”
McCabe also sees plenty of potential positives. “What seems to be common for all music creators is an interest in applying AI to help with more mundane ‘administrative’ tasks that may break the creative flow. But we’re also learning that most music creators would welcome AI as a source of musical inspiration, and we think this could take many forms.”
Learning to play, performing live, organising and searching through audio libraries, breaking through writer’s block, transforming sounds in real-time, creating new revenue streams through licensed voice models; these are just some of the areas where responsibly-built AI could benefit music creators.
Unlocking those positive potentials will be impossible without a concerted effort to rebuild consumer trust. “Overall, we would have to say that this expression of dislike for AI-generated music by the public is positive, and is an important driver of change,” says McCabe.
“But what this negative sentiment also signals is that standards and infrastructure for identifying and labelling both the presence and nature of AI in music are needed now more than ever. It is not sufficient to simply say that music is or is not made with AI – we need to describe how the AI has been used in the music creation process.”
There have been various points over the last few years where AI’s progress has felt inescapable – something that will transform our lives irrevocably, whether we like it or not. That may well turn out to be true, but the current moment is also demonstrating that people en masse do have the power to shape how this technology is developed, and to what end.
Our discontent has forced AI companies to change their tune, but only continued pressure will ensure that the new song they’re singing contains more than empty words.
