Artists Are Winning on Impersonation, Not Copyright

BySarah Fraser
ai impersonation laws: Musician recording vocals in a professional recording studio.

Right-of-publicity, contract, impersonation, and platform-labeling rules are producing the clearest early wins against AI slop right now. The sharpest court example is the July 10, 2025 ruling in Lehrman v. Lovo Inc., where a federal judge let voice actors’ right-of-publicity and contract claims proceed while rejecting the idea that voice imitation by itself is straightforwardly covered by copyright.

That split matters because it shows where the usable tools are. If the harm is “you copied my identity, voice, or account presence,” artists and performers increasingly have something concrete to point to. If the harm is “your model learned style from the internet,” the big copyright fights are still mostly active litigation, including Disney and Universal’s June 11, 2025 complaint against Midjourney, not finished wins.

A lot of anti-slop enforcement is now happening outside courtrooms anyway. Spotify, Deezer, YouTube, TikTok, Pinterest, Meta, and DeviantArt have each built day-to-day controls around impersonation, disclosure, tagging, or spam handling, not broad bans on AI content, but practical knobs that can demote, label, or remove specific kinds of abuse. That is a narrower toolbox than many creators want. It is also the toolbox that exists.

Right-of-publicity and contract claims are producing the clearest early court wins

The cleanest example is Lehrman v. Lovo Inc., a case brought by voice actors against the AI voice company Lovo. On July 10, 2025, Bloomberg Law reported that the court let claims based on right of publicity and breach of contract move forward, while cutting back the plaintiffs’ copyright theories.

That is not a technicality. It is the legal system drawing a line between copying a protected recording and appropriating a person’s vocal identity. The Loeb case summary says the court was receptive to claims that Lovo used performers’ recordings beyond the scope of their agreements and exploited recognizable voice attributes tied to identity, but it did not accept a broad theory that sounding like someone is, by itself, copyright infringement.

“The court allowed the plaintiffs’ right-of-publicity and breach-of-contract claims to proceed, while dismissing copyright claims based on voice imitation alone.”, Loeb & Loeb case summary of Lehrman v. Lovo Inc.

That is why publicity and contract claims are moving faster than pure copyright claims in imitation cases. Copyright protects particular works; publicity and contract claims can reach identity misuse and broken licensing promises. If a company cloned a performer’s voice from paid sessions or made content that plausibly passes as that performer, those theories fit the harm more directly.

The broader copyright war is still unresolved. The high-profile studio suit against Midjourney is a good example: Disney and Universal sued on June 11, 2025, and the docket shows the case remains active litigation, not a final ruling. Most of the splashy image-model cases live in that posture: important, expensive, and unfinished.

For artists, that means the practical legal advice has shifted a bit. The fastest path is often not “prove the model learned from me,” but “prove this used my name, likeness, voice, contract rights, or audience-facing identity without permission.” That is less sweeping than a landmark copyright knockout, but it is more immediately usable.

Platforms have moved faster than courts because they do not need to settle first-principles copyright doctrine to police obvious junk. They can set product rules. And the strongest current rules are mostly about disclosure, identity, and manipulation, not a blanket “no AI” stance.

A 2026 CHI paper on AI-generated content governance across social media platforms found that many systems rely on metadata, content credentials, self-disclosure, and platform-side signals rather than fully auditable detection. That is the boring but important detail: today’s AI labels are often provenance systems, not magic detectors.

YouTube has been pushing toward one clearer label format. In its update on improving AI labels for viewers and creators, YouTube said it would use a single AI-label format and new internal signals to identify altered or synthetic content. That is useful for disclosure, but YouTube framed it as a way to inform viewers, not as a universal removal system.

TikTok’s rule is similarly concrete and similarly limited. Its AI-generated content policy requires labels for realistic AI-generated or edited media and gives users reporting options. Again, the mechanism is disclosure plus reporting, not automatic exclusion.

Pinterest has gone the same route. Its help pages say Gen AI labels can be applied from metadata or by Pinterest’s own systems, and AI labels appear on Pins while content owners can identify AI-modified content. That is better than nothing, but it still depends partly on signals users cannot independently verify.

Meta has been explicit that labels are often the point. In its April 2024 policy update, Meta said it was shifting toward broader labeling of AI-generated and manipulated media, rather than removing content solely under its older manipulated-media policy. That is a transparency policy, not a slop ban.

DeviantArt is further along on creator-facing controls than many general social platforms. It offers submission labels such as Created with AI and NoAI, and its NoAI setting says the artwork is not authorized for third-party AI training datasets. That does not enforce itself across the web, but it creates a visible usage boundary artists can point to.

A policy stack is only as good as its enforcement path. Right now, the platforms with the clearest day-to-day path for creators are the ones that let them point to one of three things:

  • Impersonation of a named person or account
  • Missing disclosure on realistic AI media
  • Spam or misattribution tied to distribution systems

Those are narrower than the grand legal arguments about AI training. They are also more actionable this week.

That matters well beyond art communities. As AI tools get folded into everyday production stacks, from consumer apps to local LLMs versus ChatGPT, the enforcement problem becomes less “is AI present?” and more “is this deceptive, impersonating someone, or flooding recommendation systems?” Platforms are writing policy for that world, not for a future perfect detector.

Music platforms are building the most concrete anti-slop enforcement stack

Music services have moved further than general social platforms on AI identity abuse and spam controls. They have a stronger incentive: fake tracks, wrong-artist uploads, cloned vocals, and playlist fraud break the product fast.

Spotify now has one of the clearest impersonation rules on the books. Its impersonation policy says it will remove podcast content that impersonates another creator or host’s likeness without permission, including AI cloning. That is a direct anti-voice-clone rule, not a vague community standard.

Spotify has also built account-side controls. In its Artist Profile Protection announcement, Spotify described tools designed to keep misassigned music and AI-assisted spam off the wrong artist pages. That is the sort of operational fix artists actually need: not a debate about ontology, just “keep junk off my profile.”

Then Spotify tightened again. In a September 25, 2025 update, the company said it was banning unauthorized vocal impersonation in music and strengthening protections for artists, songwriters, and producers. That moves from profile hygiene toward identity protection in the recordings themselves.

Deezer has gone furthest on visible AI-music infrastructure. In June 2025, it announced what it called the world’s first AI tagging system for music streaming, labeling fully AI-generated tracks and describing them as a fraud risk. A year later, Deezer said its free Deezer Detector could identify fully AI-generated music with 99.8% accuracy, and that it had labeled 13.4 million tracks in 2025.

That 13.4 million figure is the most concrete anti-slop number in this stack. It says music platforms are no longer treating AI spam as an edge case. They are treating it as an ingestion problem at catalog scale.

There is still a catch. Deezer’s figure comes from Deezer, and like most platform detection claims, users cannot fully audit the underlying signals. The 2026 CHI study points to the same structural limit across platforms: provenance and internal classifiers do useful work, but they are not transparent in the way a court record or public database is transparent.

The result is an uneven map. If you are a musician or voice performer, the strongest current protections are around impersonation, misattribution, and spam distribution. If you are an illustrator arguing that a model absorbed your style, the legal path is broader in ambition and slower in results.

That split has labor consequences too. The people most exposed to synthetic substitution are often paid by task, commission, or session rather than by ownership of a large copyrighted catalog. In that sense, the current anti-slop tools line up with the same pressure visible in AI job displacement by lost tasks: the easiest thing to replace is often the small, repeatable unit of work, and the easiest thing to police is often the narrow misuse of identity around that work.

One more force is coming from regulation. The European Commission published a voluntary Code of Practice on marking and labelling AI-generated content on June 10, 2026, ahead of AI Act transparency duties taking effect on August 2, 2026. That will not settle copyright either. It will, however, push more platforms toward standard labeling and provenance habits.



The near-term verdict is plain. The anti-AI-slop wins that actually cash out today are narrow, operational, and identity-centered. Courts have been more receptive to publicity and contract theories than to imitation-as-copyright alone, and platforms are better at labeling, tagging, and removing impersonation or spam than at filtering all synthetic media.

The next milestone is not a single blockbuster copyright ruling. It is whether platform enforcement becomes more auditable as disclosure rules harden under measures such as the EU’s AI-content labeling code and August 2026 transparency duties.

Key Takeaways

  • Right-of-publicity and contract claims are producing the clearest early legal wins against AI impersonation.
  • In Lehrman v. Lovo Inc., a federal judge let publicity and contract claims proceed while rejecting a broad copyright theory based on voice imitation alone.
  • Most major copyright fights against model makers are still pending, including Disney and Universal’s 2025 case against Midjourney.
  • Platforms mostly use labels, disclosure rules, impersonation policies, and spam controls rather than outright bans on AI content.
  • Music platforms have built the strongest anti-slop stack so far, with Spotify targeting unauthorized vocal impersonation and Deezer saying it labeled 13.4 million AI-generated tracks in 2025.

Further Reading