Published August 14, 2023 · By CampaignsLive · Case Study
In April 2023, a track called “Heart on My Sleeve” appeared on TikTok, Spotify, and Apple Music. It was credited to Drake and the Weeknd. It featured the recognizable vocal qualities of both artists in a duet neither had recorded. The track had been produced by an anonymous user — eventually identified as Ghostwriter977 — using publicly available AI voice synthesis tools trained on samples of both artists’ previous releases.
The track was popular almost instantly. Within four days it had been streamed millions of times across platforms. Within a week it had been pulled by Universal Music Group, the label that holds both artists’ contracts. Within a month the broader industry had collected itself into a position about AI voice synthesis that would shape music-industry contracts and platform policies for the next two years.
The episode is in this archive because the patterns it established generalized beyond music. The questions about consent, the contractual gaps it exposed, the disclosure norms it forced into the open, and the platform-policy responses it triggered all rhyme with the conversations that brand creative production was having about AI imagery in the same period.
What actually happened, in sequence
The release. “Heart on My Sleeve” appeared without label authorization on the major streaming platforms. The track was uploaded through standard self-publishing channels, with no distinct indication that the vocal performances were synthetic. The marketing — to the extent there was any — happened on TikTok, where the track went viral first.
The recognition. Within twenty-four hours, listeners and music writers were noting that the vocal qualities were extremely close to the real artists. Within forty-eight hours, the question of whether the vocals were real had become the dominant conversation around the track. Within seventy-two hours, the consensus had landed: the vocals were AI-generated, the artists had not been involved, and the track was a high-quality demonstration of where voice synthesis technology had reached.
The response. Universal Music Group issued takedown notices to the streaming platforms and to the social media platforms where the track was being shared. Spotify, Apple Music, YouTube, and TikTok complied within days. The track was, by the end of the second week, no longer available through the major commercial channels.
The aftermath. The Recording Industry Association of America and the major labels began publicly working on policy positions about AI voice synthesis. The major streaming platforms began publishing AI-disclosure policies. Artist contracts began to include explicit language about voice synthesis rights. The Grammy organization issued guidance about AI eligibility for its awards. The cumulative response, across the industry, took about six months to fully shape and is still being refined.
The contractual gap the episode exposed
The most consequential thing the episode demonstrated was that standard artist contracts, written before AI voice synthesis was a practical concern, did not have clear language about whether the artist’s voice could be synthesized from existing recordings. The labels held the masters; the artists held the performances. Whether either of those rights extended to derivative AI-trained vocal synthesis was, in April 2023, not legally settled.
The standard position the major labels took — that voice synthesis derived from recordings the label owns is a derivative use the label can prevent — has held in practice. The legal basis remains less clear than the labels would prefer. The contractual response has been to write explicit language into new and renewed contracts that addresses the question directly, rather than relying on existing rights to extend.
This is structurally similar to what happened in advertising and brand creative around talent rights. The SAG-AFTRA AI clauses, negotiated through the latter half of 2023, did the same work for performers in commercial production that the music industry’s post-Heart-on-My-Sleeve work did for recording artists. Both responses converged on the same principle: AI use of a performer’s likeness or voice is a contractually distinct event, requires separate consent, and requires separate compensation.
The platform-policy response
The streaming platforms moved more quickly than the labels on the disclosure side. By the second half of 2023, Spotify, Apple Music, YouTube Music, and the major distribution platforms had published policies requiring disclosure of AI-generated vocal content. The policies vary in their specifics but converge on requiring that fully AI-generated vocal performances be labeled and that any unauthorized synthesis of an identifiable artist’s voice be removable on rights-holder request.
The platforms have, in practice, enforced the unauthorized-synthesis side more aggressively than the disclosure side. AI-generated music with no identifiable rights claim flows through the platforms in significant volume; AI-synthesized vocals of named artists do not. The asymmetry reflects where the legal exposure is clearest.
What brand creative production took from this
The Heart on My Sleeve episode did not directly affect brand-side advertising production, but the patterns it established found their way into brand contracts and platform policies within months. Three specific things changed.
Voice talent contracts started including AI scope. Brand-side voiceover work — the voice talent who records for commercials, brand films, internal communications, IVR systems, and digital products — increasingly came with explicit AI clauses. The clauses cover whether the voice can be cloned for future use, what consent is required for derivative synthesis, and what compensation applies. The standard position was set by union negotiations through 2023 and 2024; the non-union side caught up within twelve months.
Disclosure norms tightened in audio production. AI-generated voiceover in brand creative — whether for paid placements, internal communications, or product features — became disclosure-required across an increasing share of contexts. Some of this was platform-driven (the same platforms that require disclosure of AI-generated music tend to require it of AI-generated audio in advertising). Some was union-driven. Some was brand-side risk management.
The conversation about training data spread. The Drake/Weeknd episode made the question of “what was this AI trained on?” a public conversation in a way that the prior six months had not. The music industry’s specific concern was about voice synthesis trained on the artist’s catalogue without authorization. The broader question — what corpora were any of these tools trained on, and what rights did the rights-holders have in the resulting output — spread from music into the broader brand-AI conversation through the second half of 2023.
The pattern for what came next
The voice-synthesis pattern from 2023 has continued to shape AI policy work in adjacent categories. The visual-likeness equivalent — what happens when a performer’s face is synthesized without consent — became the central concern of the SAG-AFTRA strike. The text-equivalent — what happens when a writer’s work is used as training data without consent — became the central concern of the Writers Guild strike. In each case, the early disputes followed the same arc: high-profile technical demonstration, viral attention, rights-holder response, platform-policy reaction, contractual catch-up.
The pattern is now familiar enough that the latest generation of AI-derived content production — voice cloning of public figures, video synthesis of recognizable people, text-generation in named writers’ styles — produces faster response cycles than the 2023 cases did. The disputes are smaller because the contractual and platform infrastructure exists to resolve them more cleanly than it did in April 2023. That infrastructure is, in significant part, the legacy of “Heart on My Sleeve.”
For the related visual-likeness side of the same conversation, see The SAG-AFTRA AI Clauses. For the disclosure norms that emerged through 2023, see Marvel’s Secret Invasion AI Title Sequence.