“A safe,responsible,and profitable ecosystem of music”: Analyzing perceptions and implementation of generative AI in the music industry

Abstract

This article analyzes music industry discourses about generative AI to understand competing and conflicting views across the industrial field. Our analysis mobilizes primary data from ethnographic fieldwork collected at music trade conferences between 2023 and 2024 and secondary data from trade press, corporate statements, reports published by governments, unions, and trade bodies. Our analysis illuminates tensions and contradictions among protectionist, liberalizing, and conciliatory views toward generative AI. Some corporate actors and public stakeholders advocate for protectionist business policies and “responsible” AI development that foregrounds potential harms of AI. Other corporate actors offer more liberalizing views, encouraging investment, experimentation, and adoption of generative AI systems to cut costs and increase profits. We also note conciliatory positions, mainly from musicians’ unions and trade bodies, trying to find compromises between these two poles. We argue that these contradictions reveal a fundamentally misunderstood notion of universal AI ethics in the music industry.

Introduction

In February 2024 Universal Music Group (UMG) announced it would be removing its entire catalog from the popular short video platform TikTok. Alongside more familiar complaints about fairer compensation for artists and better copyright licensing agreements, UMG cited specific concerns over the deployment of generative artificial intelligence (AI) tools on TikTok. As the label claimed in a press release, “TikTok is allowing the platform to be flooded with AI-generated recordings . . . in a move that is nothing short of sponsoring artist replacement by AI” (UMG, 2024). The move was ultimately reversed in May 2024 after TikTok and UMG reached a new licensing deal. UMG proclaimed in a press release in August 2023 that “AI is here, and we will embrace it responsibly together with our music partners” using “industry-leading trust and safety organization and content policies” (UMG, 2023). The press release did not detail exactly what constitutes “industry-leading” protections for generative AI, leaving open the question of whether generative AI was the problem or the solution.

This struggle to shape the landscape of generative AI in music highlights issues of control in the contemporary music industries. As one of the largest music conglomerates, UMG wields considerable bargaining power to force compliance with standards it sets or risk losing access to its extremely valuable musical catalog. UMG has publicly condemned technology that might “flood the market” with AI-generated recordings while investing billions in technologies that ostensibly produce similar end results (Stassen, 2023). Other large music tech companies like Apple, Spotify, and Soundcloud have similarly invested in AI music startups to keep up with a rapidly changing system of music and technology (Robinson, 2023).

This article analyzes trends and tensions in the contemporary music industry, focusing on generative AI systems and unpacking the complex, at times contradictory and incompatible, discourse that surrounds them. However, instead of exposing these contradictions to evaluate the fitness of different approaches to the debate, we use this empirical material to shed light on a wider problem. We demonstrate that discourse and policy initiatives coming from different interest groups show the existence of misguided universalist notions of ethics in this field (Adams, 2021; Greene et al., 2019), understood as widely applicable ethical solutions to generative AI development, as we explain below. In other words, what is at stake in debates about generative music AI are wider ethical and cultural questions about the development of science and technology, power imbalances between stakeholders, and ultimately the character of human musical production as a whole. Our analysis draws on Caldwell’s (2008) notion of industrial reflexivity, whereby discourse within a business field is socially constructed through dialogue between corporate communications, discourse at trade events, and feedback from workers. While artificial intelligence has existed in some form in the music industry for decades, interest in generative AI tools specifically has surged since the 2022 launch of ChatGPT (Chow and Perrigo, 2023). Since then, industry actors, policymakers, and researchers have become increasingly concerned with the impacts of generative AI in the cultural industries. Drawing on public trade press and onsite fieldwork at international music industry trade events, we address critical media industry questions such as profitability, regulation, and power, as well as political economy issues such as fairness, crediting, and remuneration in the context of generative AI and music. In doing so, we highlight the lingering post-2000s anxieties about music tech development (Erickson, 2024) and the ongoing tension between the pursuit of commercial revenue and ideas of human creativity. We use Greene et al.’s (2019) critical approach to universality in AI ethics to point to pitfalls in these approaches to AI. We further complicate this by mobilizing Adams’ (2021) expansion of this framework for AI ethics, to highlight how these industrial contradictions are disconnected from culturally-informed perspectives on music. To conclude, we call for a more critical examination of the assumptions underlying AI ontology and ethics in the music and cultural industries than what appears in the public debates summarized here.

The next section outlines some of the existing literature on this topic and the approaches we employ to target industry culture and policy related to music AI. This is followed by a brief explanation of our methodology and its limitations. The first empirical section analyzes protectionist business policies in regard to generative AI in music and how these are articulated by major companies, aligned with or in response to governmental regulations. By protectionist, here we mean the regulatory measures oriented to secure the viability of preexisting music markets and stakeholders, such as copyright capitalization and major music industry conglomerates. In the second empirical section, we compare this to liberalizing business policies implemented by these companies. Here, we understand liberalizing as the opening of new business areas and markets for investment, as well as the welcoming of new players such as tech companies. The third empirical section outlines the perspectives of trade bodies and industry unions, which aspire to both and suggest solutions to these tensions.

Generative AI and the music industry

The use of automated systems for cultural production is far from a new phenomenon. Scholars have long puzzled over the computational creativity of machines that follow preprogrammed instructions and utilize mathematical equations to assemble literary or artistic creations (Bridy, 2012). However, a noticeable shift in recent AI music discourses is the increased focus on generative AI systems. What constitutes generative AI, as opposed to the assistive AI that the literature above refers to, is conceptually slippery. To illustrate in the context of music, LandR is a production service that offers AI-assisted audio mastering to prepare recordings to be uploaded to streaming platforms, while Suno is a platform that generates completed musical tracks from text-based user inputs. Generative musical engines are those which consist of “real time rendering software, musical assets, other backend software, and associated organizational strategies and workflows, that can be used to deliver final music tracks dynamically,” (Bown, 2024: 3). Bown explains that these engines need not necessarily rely on AI but the latter is increasingly a core component for generative music.

Scholars have traced continuity in generative AI debates that reprise discourse about the disruptive transformation of digitalization and platformization (Bonini and Magaudda, 2024). For example, we concur with Erickson (2024) that the music business continues to be deeply influenced by post-2000s culture, where stakeholders exhibit significant anxiety at being seen as techno-pessimists or losing revenue opportunities due to questioning the pace and direction of certain technological developments, as we show below. This includes that these music business stakeholders avoid supporting ideas of user control or restraining creativity. In addition, legal scholars have called attention to copyright concerns related to the data sets used to train generative AI systems (Sobel, 2018), which have become a central element for generative AI lawsuits in the music industry, as we discuss in this article. While moral standpoints about generative AI music have often been discussed in press coverage (Mulligan, 2023; Robins-Early, 2024), we seek to politicize these debates by addressing how these discourses arise through our analysis of corporate and trade body discourse in the music industry. Our empirical material shows that in the quest to reconcile post-2000s anxieties, copyright protection and exploitation, and moral valuations of generative AI, multiple stakeholders are pushing toward contradictory goals that sometimes are neither fully enforceable in practice nor serve to protect artists, as well as ignoring long-term cultural impacts and consequences of AI.

We also go beyond exposing these contradictions or legacy anxieties. Ultimately, our aim is to understand the ethical principles and specific political economies being promoted by these policies, rather than analyzing the feasibility of specific measures implemented and suggested by either side. Particularly, we focus on how the industrial contradictions that appear in our empirical material foster the emergence of notions of “responsible” AI. For this purpose, we draw on three studies more specific to AI development. We draw from Greene et al.’s (2019) analysis of the industrial culture around AI and machine learning development in computer science. They highlight the emergence of universalist notions of ethics, manifested in how discourse around AI often assumes that ethical concerns are universal, objective, and measurable, and as such can be addressed through technically informed policy. Greene et al.’s article is relevant to media studies here for its critique of the development of an industrial culture that prioritizes progress and implementation over understanding the cultural and ethical implications of these technological changes. This assumption of universality also entails believing that there is a common language to discuss these issues. As Morreale et al. (2025) highlight, the current culture of AI development relies on the flawed assumption that language (particularly English) can act as a semantic mediator for music. Because generative music systems are “reductive, exclusionary, and normalising” and cannot generate full theoretical musical models, their results will be based on cultural assumptions encoded in the software about what is important and what musical elements should AI generate. Our application of this research underlines how policy contradictions in the music industry and the subsequent development of an agenda of “responsible” AI, demonstrate the existence of those same misunderstood assumptions of–interest, class and power–universality among the different industry stakeholders, as well as their lack of engagement over wider political questions such as the implications of encoding culture into software.

Adams’ (2021) contribution bridges the relationship between this imagined shared ethics of responsibility with ideas of cultural universality. Stemming from a history of parallel assumptions about the existence of a global music market (Born and Hesmondhalgh, 2000), these ideas of universality ignore impacts on minoritized groups and regions in the industry, and how AI development in its current form deepens cultural and industrial inequalities. Adams’ (2021) approach to Western-centric morality, legality, and notions of personhood as the epistemological basis of dominant notions of ethics (p. 183) highlights how the current dynamics of AI development are based on colonial ideas of technological expansion and the universality of the English language as the vehicular language of science and development. For music and media studies, ideas of “responsible AI” and universal ethics are relevant because they contribute to a form of colonial rationality in themselves. Although we do not intend to develop an encompassing ontology of AI ethics, in what follows, we show how these grounding assumptions about AI ethics manifest specifically in the music industry.

Methodology

This study analyzes how industry representatives and insiders talk about AI, how they make sense of it, and how that sensemaking often reveals competing, contrasting, and reflexive impulses. More than simply treating AI as a site of contestation and negotiation, here we refer to the notion of industrial reflexivity advanced by Caldwell (2008: 34), which “operates as a creative process involving human agency and critical competence at the local cultural level as much as a discursive process establishing power at the broader social level.” Reflexivity is socially and culturally produced in the music industry by corporate communication, at public trade events, and in private communication with workers. This approach to industrial self-theorizing is useful because it foregrounds said flexibility, contradiction, and paradox. It recognizes the “black box” elements of the tech and music industries characterized by nondisclosure, while at the same time devising methods to shed light on these developments in ways that go beyond the fabricated urgency, despair, and hype in public press releases and artists’ statements. The empirical material below shows the contrast between artist protection, revenue expansion, and copyright control. It also addresses the tension between the capitalist principles that govern the music and tech industries and societal ideas of creativity and democratization of music.

To study industrial reflexivity, we employ a mixed-methods industry discourse analysis. Our approach blends together production studies, which foreground representation, critique, and reflexivity within the cultural industries (Caldwell, 2008), with critical media industry studies perspectives that analyze how power is constructed and negotiated between various actors and organizations who shape the industrial practices of media industries (Herbert et al., 2020). In this way, we aim to go beyond showing that generative music AI is a site of contestation to shed light on what these strategic positionings reveal in terms of cultural shifts. We collected data from two main sources. First, we collected industry discourse at music industry conferences in the European Union (including the United Kingdom) and North America. We attended five music industry conferences in 2023 and three in 2024. We attended these conferences as full delegates, collecting field observations at keynotes, panels, and roundtables. We also participated in networking sessions and social events, where we had informal conversations with managers, consultants, lawyers, developers, publishers, journalists, researchers, and artists. During and after the conferences, we also conducted interviews with industry stakeholders. Second, we collected a total of 103 English-language news articles and press releases about AI and music from major international news outlets, music industry trade bodies, and corporate communications from the official websites of music conglomerates and tech companies published between January 2020 and March 2025. We treat the information from trade press, legal actions such as lawsuits, and empirical material from industry fieldwork as different types of discourse. Conversely, we did not perform a systematic content analysis with coding categories, and instead analyzed written discourse in the same way as statements observed at industry events in an inductive manner. In this sense, industrial discourse from official statements and press releases is treated as an overall measure of the discussion, while specific statements by individuals are used to confirm it. In other words, we understand all these sources as ways of addressing Caldwell’s industrial reflexivity, or the ways in which the industry wants to present itself.

This mixed-methods approach proved effective for the purposes of our analysis, but it also entailed some challenges. The breadth and scope of this data collection allowed us to both find out about general industry trends at an international level as well as more local or region-specific business cultures. For the most part, stakeholders were open and shared industry information with us, as long as they remained anonymous. Panelists and roundtable discussants also added nuance to statements when asked informally after the presentations, so we could gather a level of detail that would not have been possible through only collecting public relations (PR) materials online. However, we are also conscious that some of the public statements of panelists and discussants follow strict internal policies from their companies or organizations, and at times, we felt that we were collecting a form of live PR discourse. Similarly, not having recordings that we could quote with precision, in the analysis below, we have prioritized written discourse to highlight the use of language about ethical concerns. However, despite these challenges in collecting discourse about technological developments in the music industry, we believe that the data presented here accurately represents current discussions in the field. We are conscious that the events we attended were limited to Europe and the United States. We were unable to collect data from important regions in the development of streaming platforms and AI technologies, such as the Middle East and East Asia. Focusing on the ways in which AI reproduces Western hegemonies (Adams, 2021) nonetheless illuminates discussions about generative AI in ways that are useful to scholars in other cultural contexts.

Protectionist discourse and public alignment

The first category of discourse we identified centers on protectionist measures and control. For context, the economic structure of the music industry is organized around the control and exploitation of music copyrights. In basic terms, revenues from recorded music are generated from licensing fees paid to rightsholders, individuals or entities that are granted legal ownership of music copyrights, by any other actors who seek to access the rights to their music. Following an intense period of mergers, acquisitions, and conglomeration of large rightsholders in the mid-2000s, more than two-thirds of popular music catalogs are controlled by three corporate entities, Warner Music Group (WMG), Sony Music Entertainment (SME), and UMG (collectively known as the Majors). The Majors command significant market power in the era of digital platforms because music service providers (MSPs) and other tech companies rely on the Majors’ licensing agreements to access valuable catalogs of popular music (Leyshon and Watson, 2025).

The Majors have backed regulatory efforts to expand their powers to prevent copyrighted material from being used to train AI systems without licensed permission, and this protectionist action was the implicit background of discussions at trade events. Here, we focus on lawsuits, legislation, and trade press coverage as extensions of industry discourse, whereby the major players present themselves as protecting the existing markets. Yet, this happens in ways that evoke a lingering post-2000s anxiety about technology development, as well as misunderstood notions of universal ethics. In October 2023, UMG, Concord Music, and ABKCO filed a suit against Anthropic, the developer of an AI model that generated song lyrics (Tencer, 2023). The court filings alleged that Anthropic violated the Digital Millennium Copyright Act (DMCA) for the unauthorized use of copyright-protected song lyrics to train its AI chatbot, Claude. One year later, the Majors issued a joint statement in support of the US Transparency and Responsibility for Artificial Intelligence Networks (TRAIN) Act (Tencer, 2024). The TRAIN Act would afford rightsholders broader latitude to take legal action if they suspected copyrighted material was being used to train AI models. Mitch Glazier, CEO of the Recording Industry Association of America (RIAA) that represents the interests of major music rightsholders in the United States, called the TRAIN Act, “a carefully calibrated bill [that] will bring much needed transparency to AI, ensuring artists and rightsholders have fair access to the courts when their work is copied for training without authorization or consent” (as quoted in Tencer, 2024).

This legal action in the United States aligns with the development of wider kinds of public policy in the United States and the EU, ¹ and the public perception of AI as a potential risk. In January 2024, policymakers in the State of Tennessee introduced the Ensuring Likeness Voice and Image Security (ELVIS) Act as a potential framework for national AI policy in the United States. Signed into law in March 2024, the ELVIS Act adds stronger protections for name, image, and likeness rights, also known as publicity rights, and criminalizes the use of generative AI to clone an artist’s voice without obtaining consent. A significant recent attempt at AI governance is the EU AI Act, a comprehensive regulatory framework designed to identify AI-related risks in Europe, which came into effect for EU member states in August 2024. The Act establishes a protocol for risk mitigation, management, and reporting, and sets compliance deadlines for AI developers. The purpose of the Act is “to promote the uptake of human-centric and trustworthy AI while ensuring a high level of protection of health, safety, fundamental rights” (European Commission, 2025: 1). However, as of March 2025, copyright-related concerns were not mentioned in any of the EU AI-Act’s risk categories. In other words, protectionist legal action set in motion by different stakeholders already presents some contradictions in their intent.

These regulatory efforts align with historically aggressive and wide-reaching attempts by the Majors to combat copyright infringement through the courts (Gillespie, 2007). For example, throughout the highly publicized crusade against music piracy in the early 2000s several large rights holders used large lawsuits or threats of lawsuits to shield their valuable catalogs from third-party peer-to-peer infringement (Wikström, 2019). In another high-profile filed lawsuit in June 2024, the RIAA announced it was filing suit on behalf of the Majors against two generative AI companies, Suno and Udio. According to the chief legal officer of RIAA (2024: 1), “these lawsuits are necessary to reinforce the most basic rules of the road for the responsible, ethical, and lawful development of generative AI systems.” Like the lawsuit against Anthropic, the Suno and Udio cases revolve around whether copyrighted material was used to train AI models. In the same line of protecting the catalog but in a more conciliatory move that did not involve a lawsuit, Sony Music Group also declared in May 2024 that it was publicly opting out of any AI development trained on its licensed content in a notice that it reportedly sent to over 700 AI development companies (Tencer, 2024).

Indeed, this existing protectionist legal action taken against AI music developers is further supported by sweeping music catalog removals. This is often expressed by large music rights holders in the press, sounding the alarm about the harms that AI music can have on artists in terms of market effects. Returning to the opening vignette, UMG (2024) temporarily removed its entire catalog of music from TikTok because it claimed “TikTok is allowing the platform to be flooded with AI-generated recordings.” In 2023, Spotify removed 7% of songs created using the generative AI platform Boomy, citing concerns about stream manipulation (Nicolaou, 2023). The exact methods Spotify uses to determine how streams are artificially manipulated are opaque; ironic in the context of a debate that places such a high premium on transparency and disclosure. In early 2025, Deezer announced its deployment of “a cutting-edge detection tool” to “increase transparency and safeguard the rights of creators,” claiming that it had already detected 10% of daily uploaded tracks to be AI-generated, and its intention to remove them from algorithmic recommendations (Bernet, 2025). However, opaque mass removals of music from popular digital platforms causenegative market effects of their own and raise the question of who benefits from this strategy. In the UMG-TikTok dispute, the suddenness of the decision left some artists worried about losing access to one of the most widely used platforms for music marketing and discovery (Beaumont-Thomas, 2024). This was especially frustrating for artists who were not signed to UMG but had, nonetheless, their music muted or removed on TikTok, because someone credited on the song was working with UMG and thus their catalogs were included in its database. These broad catalog removals do not seem to be designed with artists in mind, or at least not all of them. Indeed, this kind of 2000s-inspired lawfare strategy also contains contradictions and assumptions of universally shared interests.

The examples above show a tendency to use highly publicized actions to convey a sense of corporate responsibility. In press releases and official corporate statements, major rights holders and music tech companies position themselves as being principally focused on protecting artists and existing revenue streams. Yet, in some cases, they may enact strategies that protect their interests and the interests of large, superstar artists, while negatively impacting certain other artists. These actions often spell out the industry’s intentions to bend copyright law further in its favor. Indeed, issues with the industry’s contradictory stance toward generative AI music already appear in this protectionist discourse and action. In announcing its partnership with YouTube Music for an AI incubator, UMG stated that it supported the development of responsible AI that also guaranteed industry-leading standards of protection and control, therefore paradoxically continuing to permit AI music to “flood the market.” UMG CEO Lucian Grainge stated,

Our enduring faith in human creativity is the bedrock of Universal Music Group’s collaboration with YouTube on the future of AI. Central to our collective vision is taking steps to build a safe, responsible and profitable ecosystem of music and video – one where artists and songwriters have the ability to maintain their creative integrity, their power to choose, and to be compensated fairly. (UMG, 2023, emphasis added)

In the case of SME’s content notice letter, the opting out of AI development equally acknowledges that it and its affiliates, “. . . have been embracing the potential for responsibly produced Al to be used as a creative tool, revolutionizing the ways songwriters and recording artists create music,” adding that “innovation must ensure that songwriters’ and recording artists’ rights, including copyrights, are respected” (SME, 2024).

On the streaming side, Spotify also remains bullish about hosting AI music, and the removals of music outlined above happen in parallel to an espoused acceptance of AI-generated content. As of December 2024, Spotify had no explicit policies banning AI music. When asked in a public interview about whether generative AI was “good for Spotify” the company’s co-president, CTO, and CPO Gustav Söderström responded, “. . . if creators are using these [AI] technologies – where they are creating music in a legal way that we reimburse and people listen to them – and are successful, we should let people listen to them,” (Kantrowitz, 2024).

Discourse about “responsible AI development” was a recurring theme in trade press and promotional communication we collected, and a striking example of contradiction and universalist ethics assumptions. AI incubators founded by or in partnership with music companies and AI startups acquired by music tech claimed to be developed responsibly (Ogul, 2024; RIAA, 2024; UMG, 2024). In other words, AI development is framed as “responsible” when it is being developed in partnership with large music companies. This makes sense at first glance because music companies understand the industry and claim to represent the interests of artists. We interpret these initiatives as protectionist because, in practice, “responsible AI” appears to mean expanding control of the burgeoning AI music market, using automated tools that give rightsholders more power to monetize and police their copyright assets across digital space at scale. These tools are framed as being deployed to help artists get paid, despite issues with automated content moderation identified in previous research (see Gray and Suzor, 2020).

From the evidence above, we draw two insights. First, discussions around protection against AI invoke a form of industry culture akin to the early 2000s lawsuit strategy against streaming platforms, and highly publicized actions to invoke corporate responsibility, again raising the question of who benefits from them. Second, the discourse on corporate responsibility also seems to evoke a kind of moral universality (Greene et al., 2019) and implies that ‘ethical AI’ development is in the interest of every stakeholder involved. However, these initiatives show that even different divisions within the same company may have incompatible interests. Indeed, this essential contradiction between aggressive lawfare against AI developers and the push for “responsible AI” partnerships points to the next type of discourse we address here.

Liberalizing discourse and industry alignment

The second category of discourse that we observed at international industry trade events espoused clear enthusiasm toward embracing generative AI and liberalizing the market, often in an informal way, hidden under the guise of debating these technologies. This discourse appeared in ways that seem to be more directly targeted toward industry insiders than the widely reported cases outlined in the previous section, evidently with the intention to align with industry interests rather than public concerns, yet still evoking the kind of shared preoccupation about AI ethics and corporate responsibility of the previous section. Returning to the lawsuits highlighted, when viewed alongside these liberalizing approaches to AI, it becomes clear that large rightsholders in the music industry indeed support the expansion of generative AI, provided that key players remain in control and AI ventures can be monetized. On a panel at a large trade event in the United States, representatives from an MSP, a label, and a publisher debated how the kinds of protectionist approaches described in the previous section could pave the way for new business opportunities:

MSP:

“We don’t want to limit creative tools, but the industry needs to have a better handle on AI policy. It needs to have the same treatment as copyright”

Label:

“We are excited about the potentials of [generative AI]/ now is the scramble to streamline things and test it out”

Publisher:

“We have many teams trying to figure out how to use it”

Exchanges about AI at trade events that took place before 2024 were often accented with vague appeals to policy and governance via existing copyright doctrine or copyright reform. In some cases, however, it was hard for panelists to mask their excitement about the potential for AI to help circumvent issues that arose from working with human artists. Regarding licensing AI-generated content, on a different panel at the same US trade event, an entertainment lawyer encouraged clients to use AI generative soundalikes if licensing a particular song or artist is too burdensome or expensive. As an exception, on this occasion another panelist who represented a music clearance company pushed back, reminding that soundalikes were usually not acceptable workarounds for licensing, asking why AI should be any different. Paradoxically, the lawyer had assumed the liberalizing stance while the clearance company had taken a protectionist one.

Confirming this ambivalent discourse about ethical concern and liberalizing AI development, Chris Horton, senior vice president of strategic technology at UMG, explained how the protectionist and liberalizing approaches are two sides of the same coin:

Development will continue at pace, but I hope that, just like in the P2P era, licensing, litigation, and legislation will eventually align so that all AI companies see that cooperation with creative industries is the obvious path forward. This will unlock new investments and businesses, leading to even more products and services (Dredge, 2024)

The same discourse came from streaming companies. In line with Söderström’s statement above, Spotify CEO Daniel Ek explained in an interview that he had no plans to prohibit AI music from the platform (Kleinman, 2023). In addition to having no plans to ban AI content in its catalog, low-cost, AI-versioned stock music known as “Perfect Fit Content” is indeed an important area of business development for Spotify in its quest to generate profits (Pelly, 2025).

This liberalizing discourse appeared alongside waves of investment into Gen AI music startups and development in 2023–2024 (Stassen, 2023). Suno used contradictory statements to justify this investment, claiming to be both designed for professional musicians and to be oriented toward public users:

Our community of musicians deserve the very best tools.

While GRAMMY-winning artists use Suno, our core user base consists of everyday people making music – often for the first time. (Shulman, 2024)

Besides these investment waves and PR statements, we were struck by the ways discussions about liberalizing AI development were often trivialized at trade events. Once again, the post-2000s culture among industry stakeholders wanting to make clear that they were not opposed to technology reappeared. A significant part of the pitching sessions in these events were reserved for AI-based startups. In a rather fickle structure, these sessions were scheduled alongside debate-style panels announced as critical AI perspectives. But debates around generative AI often revolved around how to use it, without any critical reflection. The contrast between the interests of streaming platforms, rightholders, established artists, and newcomers was also evident, and we again thought that a universal ethics of “responsible” AI development was not present in, and could not arise from these discussions.

Multiple trade events also turned to celebrity culture to generate social media attention by simply announcing that there was a discussion of AI. Yet, famous artists, producers, and journalists scheduled as panelists generally did not offer more complex analysis than members of the audience, contributing to a kind of celebrity worship atmosphere. We recall a particularly poorly managed panel where several famous musicians and producers (incidentally, all white men) were scheduled to discuss AI developments, and where each panelist seemed to produce a more sensationalist statement than their peers. One musician claimed that generative AI would reach the point of perfection that would make humans irrelevant. However, he said that this did not concern him since he was already a millionaire. In response, a producer compared the current stage of AI development with the early days of the Internet (i.e., early 2000s), as something risky but ultimately worth investing in, without any specific suggestions. A third artist warned about the market risks of AI for musicians, but without suggesting any countermeasures. Overall, the social media impact of these sessions seemed to be the objective, more than serious industry reflection or planning. In turn, we could see those very statements being disseminated by the audience on social media, notably via professional networks on LinkedIn, further spreading this celebrity authority and superficial approach to the issue.

From the evidence above, we note three insights. First, the unresolved tension between AI understood as a new form of revenue creation on one hand, and as a creative tool on the other, dominates these dialogues. Second, here we see again confirmation that a perceived universality of ethical concerns about AI (Greene et al., 2019) is not genuine in practice; the objectives of venture capitalists, tech developers, established artists, independent musicians, public users, and government regulators can hardly be approached by one single form of intervention. Third, the discourse around generative AI seems polarized between hardcore protectionist lawfare and liberalizing markets and promoting investment. The next section outlines proposals advanced by trade unions and artists, which show alternatives between these two protectionist and liberalizing poles, yet are still plagued by this misunderstood ethical universality of ‘responsible’ AI.

Fairness discourse and labor alignment

Amid these contradictory industry attitudes toward generative AI are the perspectives of musical artists. Although musicians have expressed opinions in individual public statements, here we focus on the views of artists as represented by industry trade bodies, advocacy groups, and professional unions and organizations we observed during our industry fieldwork and trade press. There is some overlap between the discourse from trade bodies and corporate communication, particularly around issues such as consent, copyright, remuneration, and transparency, without the same emphasis on turning profits and techno-optimism. Similarly, there is some overlap between discourse from trade bodies and corporate communication, emphasizing monetization of the catalog.

The first issue for artists and musicians is that of informed consent. In September 2023, the Council of Music Makers (CMM), a UK-based group representing five music trade bodies, released a statement on AI fundamentals to guide fair policy and industry. Leaders from the CMM discussed these fundamentals at a trade event we attended in 2024. Obtaining consent from artists was the top priority for the CMM. They demanded that AI developers consult musicians for any AI projects or tools that use artists’ copyrighted music or likeness; that AI-generated musical outputs created with consent from artists must always be labeled as such; and, crucially, that consent must be obtained explicitly and not inferred by rights holders or technology companies (CMM, 2023). In turn, artist organizations’ emphasis on consent and respecting the name, image, and likeness of artists points to further issues, namely the exploitation of publicity rights for generative AI. Even when ethical practice and explicit consent are sought, AI lookalikes developed by companies like UMG may still harm upcoming artists competing in a market with AI clones. In addition, we note that consent being offered on behalf of artists by rights holders is a likely point of contention amid the AI-development projects and partnerships detailed in the previous section. Consent in the music business can often border on coercion, given the significant and historic power imbalances between rightsholders and artists (Stahl, 2012). Here, we reiterate that once the priorities of artists and musicians at different career stages are considered, a notion of shared ethical concern among stakeholders in the industry is even less present. Yet, these demands indicate that trade bodies believe such an agreement is possible.

A second concern highlighted by several trade bodies was copyright protection for human musicians. Two strategies are noteworthy here. First, some trade bodies call for restrained approaches to policy reform. The Association of American Independent Musicians (A2IM), a US-based trade body, published a set of core principles in AI in March 2023. Like the CMM, the A2IM urged AI developers to exercise compliance when using copyrighted works for training AI models and called for a licensing process for musicians who wish to opt-in. Unlike the CMM, the A2IM (2023) warned that “governments should not create new copyright or other IP exemptions that allow AI developers to exploit creators without permission or compensation,” joining calls for informed consent while advising against expanding copyright enforcement powers. Second, other artist organizations call for regulatory action. The International Federation of Musicians (FIM) highlighted how EU copyright policy fails to account for AI. According to their statement, addressing copyright questions means adapting legal systems and frameworks to AI-generated music (FIM, 2024). It does not mean rolling out takedown systems that afford blanket power to the Majors and tech companies. The FIM stresses that the EU AI Act does not adequately address musicians’ concerns because it does not include copyright infringement as a potential risk. They suggest remedies to prevent the displacement of human artists and to ward against market distortion. It is also important to note that these fora downplay alternatives to copyright enforcement. For example, in the United States, protections could come from stronger publicity rights, as the ELVIS Act seeks to establish. However, this was completely absent from these discussions. Once again, the issue of copyright infringement highlights how the majors, artists, and lawmakers may have different ethical concerns that require different policy responses, but their actions suggest that they believe to have shared interests and a common notion of universal ethics.

The third issue highlighted by trade bodies pertains to remuneration. The International Confederation of Societies of Authors and Composers (CISAC) published a report in December 2024 expressing several concerns about AI music “cannibalizing” artist revenues (CISAC, 2024). Though CISAC predicts a 21% loss of recorded music revenues due to generative AI by 2028, Björn Ulvaeus, president of CISAC and founding member of ABBA, said

There is of course no way we can or should stand against AI. We’re not Luddites. AI can be and is a wonderful tool, and is already being used in all kinds of exciting ways . . . But this progress must never be at the expense of creators’ rights . . . and ensuring fair remuneration . . . (Tencer, 2024)

Worthy of note is that Ulvaeus is also one of the partners of the UMG x YouTube AI incubator and one of the loudest voices supporting Gen AI music. His involvement in the industry as a trade body representative, legacy artist, and AI investor exemplifies well the contradictions that we have discussed above, and the lack of a universal ethics of AI in practice. Even the same person can perform multiple stakeholder roles in the music and tech industries, and each one of these can have opposing ethical implications. We see Ulvaeus as a clear instance of an individual invested in the post-2000s anxiety of losing investment opportunities, while at the same time representing other artists’ concerns about revenue collection, and also corporate concerns around copyright enforcement.

In relation to remuneration, the fourth issue underlined by trade bodies and artists is transparency. At a 2024 conference in the United Kingdom, representatives from the CMM highlighted how transparency, metadata improvement, and remuneration are crucial in the digital market. The metadata summit at another trade conference we attended in 2023 also emphasized issues of transparency about how streaming services work and distribute revenue. Barriers to releasing accurate metadata to and from streaming platforms were highlighted as impacting revenue collection significantly, as they are the basis of correctly crediting artists. We agree with these trade bodies in that a crucial problem in the development of generative software is how little consideration tech companies and the music industry have given to systematically coding metadata (Campos Valverde 2025) and how this affects financial aspects of the digital music market. Transparency and remuneration are indeed two sides of the same coin, since metadata biases affect how the musical catalog can be read by the software behind generative AI, therefore increasing monetization inequalities. A catalog with low-quality, sparse data points and little granularity is not searchable, readable, and therefore not monetizable, whether by streaming or generative AI software. However, we also noted how industry panelists often blamed the lack of training, particularly for indie and DIY artists, as responsible for metadata inaccuracies when creating and distributing works in the first place. In the first panels we attended in the United States and the United Kingdom, it was often mentioned that training independent artists or removing them from the value chain was the best solution. We noticed the lack of shared ethical values between the transparency the artists demanded, the concerns about monetization and exploitation of content, as opposed to crediting, and suggestions to restrict access to artists. Once again, it is unlikely that the ethical principles of transparency that musicians seek to redress are shared by tech companies and their understanding of proprietary data; the ethical objectives of both stakeholders are simply diametrically opposed.

To the trade bodies’ focus on this deficient metadata workflow and uneven monetization in the music and tech industries, we would like to highlight a last point that is further reproduced by AI. There is a more cultural question in regards to the approach to technology development, which also points to a wider cultural problem with universalist understandings of ethics (Adams, 2021). Important metadata elements in the crediting of artists and collection of revenue, such as default language, are designed from an Anglo-centric perspective, and there are significant existing biases in favor of Western music (Campos Valverde 2025). However, most of the industry and these trade bodies understand software design as a descriptive endeavor that can be delegated to tech developers. Instead, we contend that metadata coding is a prescriptive practice and a generative technology, since it creates categories and hierarchies of music. Yet, to our knowledge, no music or tech companies have addressed inequality issues in the technical fabric of generative tools, even when the problem is openly admitted and discussed in trade events. The metadata summit mentioned above was a relatively more engaged exception in this respect, while the issue was simply superficially acknowledged in passing at all the other events that we attended, academic conferences included. In terms of actionable initiatives to improve metadata, the metadata code published by the UK Intellectual Property Office compels all parties in the value chain to submit accurate artist, songwriter, work, and recording metadata for each release (UK IPO, 2023). Similarly, the UK code of practice on transparency aims to improve access to information about royalties and licensing (UK IPO, 2024). However, representatives from the CMM pointed out at a trade event in the United Kingdom that this code does not require digital service providers like streaming platforms to release information proactively and requires opt-in from large rights holders. It establishes a mechanism to lodge complaints, but no real means to ensure compliance.

This discussion of metadata and monetization is even more politically relevant when it comes to the development of generative AI, because it is plagued by assumptions of a universality of ethical concerns that does not exist in practice. It is not just that artists and industry stakeholders do not focus on the same issues; industry and artists in different countries or genres are differently affected by culturally insensitive biases in the training of AI. The industry’s focus on copyright and investment and artist organizations’ focus on remuneration misplace the core issue, which we argue should be the underlying assumptions about what constitutes ethical technology development, particularly technology developed for cultural production, and how to agree on principles of best practice. These bodies are right to think that issues of inequality and monetization are intrinsically related to the management and transparency of technical aspects of these systems. In systems already biased toward Western music and culture, however, the development of generative AI for music is likely to further reproduce preexisting cultural hegemonies. The imposition of English language-centric systems and using limited musical models that rely on flawed cultural and computational assumptions about music will only exacerbate existing biases (Morreale et al., 2025).

This is particularly striking, for example, when comparing the limitations of Western notation to represent the music of large regions of the world, such as the Middle East or the Indian subcontinent. Moreover, language and semantics are plagued by cultural and power inequalities that cannot be eliminated from coding, particularly when the default vehicular language of software development (and science more broadly) is English. Software developers (and derivative product designers such as metaverse music performance and gaming) admitted at public trade events that they know that generative AI is not designed for any music outside the West; they simply expect that AI companies will catch up at some point.

Here we concur with Adams (2021) that AI development is not just biased in its technical construction; it is biased in its fundamental epistemology. The assumption that the generation of music can be semantically coded and prompted is already a core ontological problem, and a particularly Western-centric vision of music as something that can always be transcribed into written form (as opposed to an aural experience and transmission). The discourses around AI that we have presented in this article not just contribute to showing this flawed universalist understanding of ethical concerns related to AI development between different stakeholders, but also emphasize how universalist notions reinforce the idea of the West as the center of rationality and scientific advancement:

. . . a central assumption within AI: that intelligence and the production of knowledge can be outsourced to a machine presupposes such knowledge to be both separable from the context in which it was produced and applicable to other contexts and realities. (Adams, 2021: 185)

What is ultimately at stake in debates about generative AI in music is thus something far deeper than the conflicting interests of industry, rightholders, artists, and software developers. Why should music as a whole be encoded in words, particularly in the English language, and why should it be coded with Western music terminology? Why should tools developed in the West be considered universally applicable scientific methods? Every one of these agents has fallen short of considering questions that even computer engineers are still debating, about whether the artistic potential of generative AI in music “might lie beyond attempts to replicate human music-making methods” at all (Morreale et al., 2025: 300). In other words, the discourses about AI that we observed show the increasing influence of tech-centric thinking creeping into the music industry.

“Safe, responsible, and profitable” ecosystems of music

It comes as no surprise that the profit-driven imperatives of corporate entities are at odds with individual musical artists and the trade bodies representing them when discussing AI. However, the reflexive attitudes expressed within corporate entities reveal internal inconsistencies that arise from trying to have it both ways. On one hand, large rightsholders urge caution and restraint when dealing with technology they deem to be “risky.” They point to the ways that AI threatens the livelihoods of musicians and support efforts to regulate AI, provided the policies tighten controls they hold over these emerging markets. On the other hand, many of these same entities are heavily invested, financially and symbolically, in developing generative AI systems. This is likely driven, in part, by a post-2000s anxiety of trying to be ahead of the technological curve. It also implies that the inherent risks of AI are permissible when the technology is developed “safely and responsibly” and builds toward a “profitable ecosystem.” As the evidence presented here shows, safety and responsibility are hugely subjective concepts. This contrast between responsible and profitable is apparent in statements from music industry trade bodies seeking expanded informed consent, stronger protections of human creative expression, fairer remuneration, and improved transparency.

We believe that exposing these contradictions and anxieties is important; yet, here our contribution is to push debates further toward understanding the reflexive dynamics behind these policies and perspectives. Value-laden platitudes expounded in industry discourses perpetuate misunderstandings and assumptions about universal ethics. Within the various competing viewpoints toward music AI presented here, lies the belief that some unified version of a “safe, responsible and profitable,” or in other words, “ethical” AI exists for all these differently positioned and competing stakeholders. These contradictory policies also show that these sets of actors believe that the impact of such universal ethical AI is attainable and measurable via financial or datafied instruments. Moreover, this distorted appraisal of universally applicable ethics further reinforces the idea of a detached, neutral, and objective rationality that can be disconnected from the place where it is produced. The evidence above demonstrates that such a neutral ontological space does not exist, even within the same area of the industry, let alone among differently positioned cultural markets. In addition, as Adams (2021) notes, the notion of this objective intelligence is itself an illusion, sustained by a colonial interpretation of knowledge creation and technology development. Critical debates around AI in the music industry will not move forward until these antagonistic perspectives and the illusion of universal ethics are acknowledged and more directly addressed by researchers, policymakers, and other key figures within this industrial field. This may be difficult for the cultural industries as they increasingly embrace the opaque and unregulated culture of the big tech industry and the launching of software that is still under development. Future research should continue to investigate the cultural and economic complexities of generative AI adoption in the music industry.