On an ordinary spring day in 2023, a song shook the music industry to its core. It wasn’t the lyrics or melody, those were catchy enough, but the fact that the artists supposedly performing it had nothing to do with its creation. The track, titled “Heart on My Sleeve,” sounded uncannily like a collaboration between megastars Drake and The Weeknd. In reality, it was the handiwork of an anonymous TikTok producer called Ghostwriter977, who used an AI trained on those artists’ voices and styles to generate the song. Millions of fans streamed the eerily convincing fake. For a moment, it blurred the line between human artistry and algorithmic mimicry. But that moment didn’t last. As the track went viral, the legal alarms at Universal Music Group, Drake and The Weeknd’s label, rang loud. Within days, platforms like TikTok and Spotify had taken the song down in response to Universal’s copyright complaints. The “Heart on My Sleeve” saga is more than just a music industry anecdote. It’s a harbinger of the legal battles that have since moved from message boards into courtrooms and statute books.
Three years on, we finally have some answers, or at least court rulings and legislation to reference instead of pure speculation. This article now reflects where the law actually stands in 2026.
Ghosts in the Machine: Creativity Meets Complexity in Court
Ghostwriter977’s AI-generated hit raised questions nobody could ignore. Was this unauthorized use of Drake’s and The Weeknd’s voices a form of theft, or just clever homage? Who owns a song with no human singer, and could the song itself earn copyright protection? U.S. copyright authorities have already staked out a position that purely AI-created works, lacking a human author, cannot be copyrighted. But the flip side is thornier: if an AI-generated track leans heavily on a real artist’s style or training data, could that violate the artist’s rights? Universal’s takedown of “Heart on My Sleeve” was enabled by a small detectable sample in the track, a producer tag embedded in the AI’s training material, that gave the label a clear copyright hook. Had the AI been more precise and left no trace of the original recordings, the legal basis for removal would have been much shakier.
This ambiguity reveals a deeper problem: not just what AI copies, but who gets to accuse and enforce it. Independent music creators, particularly those experimenting with AI tools, have faced false copyright claims on platforms like YouTube. These are not remix pirates or impersonators. They are original creators livestreaming their production process, only to receive takedown notices from algorithmic enforcement systems. One common story involves a creator being flagged by Content ID for their own track, because an AI-generated model triggered a false positive or matched improperly catalogued training data.
As one artist put it in a YouTube discussion about this growing problem: “I’ve been copyright claimed several times on videos of recorded livestreams of me creating the actual tracks that I’m being claimed with. It’s a disaster.” Other creators describe being flagged for synthetic vocals that sound like someone else’s, even when generated legally or with licensed models. The root issue is that big tech platforms have deployed automated copyright enforcement systems like Content ID that prioritize claimant protection over creator due process. These systems operate at massive scale, yet rarely allow real-time rebuttal or human review before a strike is issued.
This ecosystem fosters a kind of platform-driven copyfraud, where companies or third-party firms file automated claims, creators lose monetization or visibility, and appeals drag on for weeks. The legal deck is stacked: under current DMCA provisions, platforms are incentivized to act on takedown requests immediately to maintain safe harbor status, even when claims are baseless. Legal commentators have started calling this pattern “Copyfraud 2.0,” a phrase that captures how copyright law gets used to suppress creativity rather than protect it. For AI audio creators, this adds another layer of vulnerability. Even those using properly licensed synthetic voices, or models trained on their own vocal data, face a real chilling effect from false flags and wrongful takedowns. In an age where a voice can be both a medium and a liability, platform policy is increasingly as important as legislation.
Black Box Algorithms and the Fight for Accountability
The music industry is no stranger to legal innovation, or to legal inertia. It now finds itself at a critical point where AI-generated vocals, algorithmic attribution, and automated enforcement mechanisms are colliding in unprecedented ways. That collision is a battle not only over rights and royalties, but over the very concept of authorship and identity in an AI age.
Digital distribution services, labels, and streaming platforms increasingly rely on black-box AI systems to assess copyright claims, match audio fingerprints, and determine royalty allocation. Their decisions are often made without explanation, and in many cases, without human review. Artists, producers, and publishers have started questioning the legitimacy of these opaque processes, especially when false claims lead to lost income, copyright strikes, or shadowbans.
Independent label Concord Music Group sued an AI audio-matching service deployed by a major distributor in 2024, alleging that their artists’ earnings were misdirected due to algorithmic errors and that attempts to challenge the allocations were ignored. The complaint cited negligent automation practices and a violation of contractual audit rights.
Legacy organizations have been forced to adapt too. The Recording Academy and the Music Publishers Association have both convened task forces on AI ethics and attribution, urging lawmakers to require explainability and auditability for AI systems that determine revenue flows. Organizations like SAG-AFTRA, ASCAP, and BMI have lobbied for provisions that keep human creators in control of royalties, even when AI is used to mimic or remix their work. Universal Music Group’s “Voice Rights” division, created in 2025, now reviews contracts for AI clause inclusion, spelling out exactly who can use an artist’s vocal likeness and under what terms. Warner Music has piloted watermarking systems that embed metadata into AI-generated vocals so they can be traced and verified.
The 2026 Update: When the Courts Caught Up
If 2023 was the year the industry noticed the problem, 2026 is the year the problem got a legal answer, or several partial ones.
The clearest signal came from Germany. On July 31, 2026, the Munich Regional Court ruled against Suno in a lawsuit brought by GEMA, the German music rights collecting society, over six compositions, including works by Falco and Rammstein, that Suno’s v3.5 and v4 models could reproduce in near-recognizable form. Suno argued its models learn statistical patterns rather than memorizing specific works, and that this should fall under Germany’s text-and-data-mining exception or U.S. fair use doctrine. The court rejected both defenses. Because the original compositions remained extractable from the model as outputs, the judges treated this as memorization, not abstraction, and memorization counts as reproduction under German copyright law regardless of how the model got there.
The U.S. case is running on a similar track. RIAA’s suit against Suno is still active in federal court in Boston. Warner Music settled in late 2025 in exchange for a licensing partnership, artist opt-in controls over voice and likeness, and Suno’s acquisition of Songkick. Sony and Universal have not settled. In May 2026 they moved to expand the case from 560 tracks to more than 61,000 after discovery showed how much of their catalogs had gone into training, a move that could push potential damages past nine billion dollars. Suno is fighting that expansion. Read together, the German and American cases point in the same direction: courts are losing patience with “it’s just pattern-matching” as a defense once a model can be prompted into reproducing something identifiable.
Regulation is catching up too. The EU AI Act’s transparency rules, under Article 50, became enforceable on August 2, 2026. AI-generated or manipulated audio, including deepfake voice content, now needs machine-readable markings so platforms can detect and label it. General-purpose AI providers must document their models, publish training-data summaries, and maintain a copyright policy. This applies to any company placing an AI system on the EU market, so it reaches Suno, Udio, and any voice-cloning tool used by artists working with European labels, distributors, or collecting societies.
In the U.S., the NO FAKES Act is in its fourth congressional iteration and, by most accounts, has its best chance yet of becoming law. The 2026 version would create a federal property right over a person’s voice and visual likeness in digital form, with licensing mechanisms and DMCA-style notice-and-takedown enforcement, plus a new counter-notification process so someone accused of misusing a likeness can contest a takedown. It has bipartisan sponsorship in both chambers.
State law already offers some protection. Tennessee’s ELVIS Act, in force since 2024, made voice a protected property right and specifically targets tools built to clone it without consent. It has already been tested: after producers used an AI tool to replicate singer Jorja Smith’s voice for a track that went viral, her label issued takedowns and demanded royalty shares, and the track was pulled from streaming until it was rerecorded with her actual vocals. Courts outside the U.S. have reached similar conclusions. India’s Bombay High Court ruled in 2024 that unauthorized voice mimicry violates an artist’s personality rights, a precedent now cited well beyond Indian courts.
None of this has solved Copyfraud 2.0. Independent creators still get hit with algorithmic false copyright claims, sometimes on footage of their own recorded livestreams. More AI content in the pipeline means more noisy signals for automated systems to misjudge, not fewer.
Why Responsible AI Matters to Us
At Sonarworks, we believe AI should empower creativity, not replace it. That’s why we’re proud to support the Principles for Music Creation with AI, joining industry leaders committed to the responsible development and use of AI in music. These principles promote transparency, respect for creators’ rights, and the belief that technology should amplify human creativity while protecting the people behind it. It’s an important step toward building AI tools that creators can trust.
That commitment shapes how we build our own products, not just how we talk about the industry. Here’s how our product owner, Anatolijs, describes it:
“We’re proud to have our own proprietary methodology for how we create and train voice presets. All presets are created by working closely with real artists. Every artist we work with is fairly compensated, and they give us their rights so creators can use presets trained on their voice for future royalty-free creations. That’s also exactly why our preset library isn’t huge. It’s intentionally limited. Every preset we offer was obtained legally and with full consent, and we’re not going to inflate that number by cutting corners on either.” – Anatolijs, Product Owner at Sonarworks.
What This Means If You Make Music for a Living
If you’re a songwriter, producer, or audio engineer working with AI voice or composition tools, the rules have gotten clearer, even if they’re not fully settled. A few habits are worth building into your workflow now, before a rights dispute forces the issue.
Document consent and provenance, not just for yourself but for anyone whose voice, performance, or likeness ends up in your session. If you use a voice preset, a cloned vocal, or an AI-assisted composition tool, keep a record of the license or terms of service that governed it, the date, and what it explicitly permits. Royalty-free commercial use is not the same as “you can post it,” and neither guarantees you can sell it into a sync license. Courts and collecting societies are increasingly asking for exactly this kind of accountability chain, from data acquisition through to output, and “the platform said it was fine” is a weaker answer than a saved license agreement.
If your track uses AI-generated or AI-modified vocals and it’s going to European distributors, platforms, or collecting societies, assume it needs AI-content labeling under Article 50 of the EU AI Act. That labeling usually means machine-readable metadata rather than an audible disclaimer, so it’s as much a distribution and metadata question as a creative one.
Don’t rely on fair use or the text-and-data-mining exception as a shield if you or your studio are training or fine-tuning your own model. The Munich court’s reasoning, that a model reproducing extractable, recognizable fragments of a copyrighted work counts as reproduction regardless of training method, is likely to influence how other jurisdictions read similar defenses. If you’re training custom voice or instrument models on copyrighted recordings without a license, that exposure is real and growing.
Keep an eye on the NO FAKES Act, but don’t wait for it to pass before protecting yourself. Until federal protection exists, protection is a patchwork of state laws, with Tennessee’s ELVIS Act the most tested, platform terms of service, and your own contract language. If you’re licensing a voice for a project, human or AI-trained, put the same specificity into that clause that you’d put into a sample clearance: what’s allowed, for how long, in what territories, and whether it covers derivative or AI-trained use at all.
Push back on Copyfraud 2.0 claims, but expect to have to. If your original, human-performed work gets flagged by an automated system that mistakes it for AI-generated or infringing content, the claim probably won’t correct itself quickly. Keep session files, stems, and timestamps as your evidence trail. The same discipline that protects you in an AI dispute protects you here too.
Treat royalty-free AI voice presets as a spectrum, not a guarantee. The safest presets are the ones where you can trace the chain back to a named artist who consented and was paid, which is a deliberately narrower field than every preset library out there. A smaller, verifiably-licensed set of voices holds up in a dispute in a way an unlicensed, unlimited one won’t, no matter how good it sounds in the mix.
For sync, licensing, and commercial placements specifically, ask the question before you sign, not after. Buyers, labels, agencies, streaming platforms are increasingly building AI-provenance checks into their intake process. Being able to answer whether any voice or performance in a track was AI-generated or AI-assisted, and if so, whether it was licensed, cleanly and quickly, is becoming a basic professional expectation rather than an edge case.
Rewriting the Rules of Creativity
The success of AI in music will be measured by whether it strengthens or erodes the structures we’ve built to protect creative labor. One case still illustrates this tension well: when Warner Music discovered AI-generated tracks replicating the vocal likenesses of its frontline artists, the label didn’t just issue takedown notices, it launched a top-down internal audit. Warner’s General Counsel called for a systemic review of how vocal identity is protected in licensing contracts going forward, stating plainly that “innovation must always respect human artistry.” Warner followed up with a new internal policy requiring all A&R and licensing contracts to address potential future uses of AI voice models, even where none are currently planned. Other major labels have since started mirroring that stance.
We are the architects of this future, not just as developers, performers, or executives, but as a global community of listeners and creators. It is up to us to decide whether AI serves as a tool for empowerment or exploitation. Ethical frameworks, legal innovation, and business strategy all matter, but our collective values have to lead. We have to resist the temptation to offload accountability to machines and instead design systems that protect artistic expression, preserve human dignity, and reward contribution fairly.
Continue reading our blog to learn more.
