Vondran Legal® - IP and AI Voice Misappropriation Update on Google Case.

Case: Greene v. Google LLC and Alphabet Inc., No. 3:26-cv-01773-CRB
Court: United States District Court for the Northern District of California
Judge: Hon. Charles R. Breyer
Decision: July 23, 2026
Result: Plaintiff's motion to remand granted

Can an AI Company Imitate a Famous Voice Without Permission?

A federal district court has issued an important early ruling concerning artificial intelligence, voice imitation, copyright preemption, and California's right-of-publicity laws.

In Greene v. Google LLC, public radio and podcast personality David Greene alleges that Google's NotebookLM artificial-intelligence product uses a synthetic male host whose voice sounds substantially similar to his. Greene claims that the AI host—identified in his complaint as “Charlie”—reproduces his sound, tone, cadence, and vocal mannerisms without authorization.

Google removed the lawsuit from California state court to federal court, arguing that Greene's claims were effectively copyright claims because they depended on the alleged copying of Greene's recorded radio and podcast performances.

The federal court disagreed.

Judge Charles R. Breyer held that Greene's claims principally target the output of Google's AI product—the allegedly unauthorized imitation of Greene's voice and identity—not merely the input, such as the possible copying of sound recordings to train an AI model. Because a person's voice and identity are not themselves copyrightable works, the court concluded that Google had not established federal copyright jurisdiction.

The court therefore returned the case to California state court. Importantly, however, the court did not decide whether Google actually copied Greene's recordings, whether NotebookLM's synthetic voice legally misappropriates Greene's identity, or whether Greene will ultimately recover damages. The decision addresses where the case should be heard, not who should ultimately win.

The full remand order is available through the federal court filing published by Justia.

Who Is David Greene?

David Greene is a nationally recognized radio and podcast personality who has hosted and moderated several National Public Radio programs. According to his complaint, his broadcasts have reached millions of listeners and made his voice widely recognizable.

Greene alleges that he has no business relationship with Google or NotebookLM and never authorized Google to use or imitate his voice.

According to the complaint, journalists, former colleagues, friends, and NotebookLM users began asking whether Greene had licensed his voice to Google. Greene alleges that NotebookLM's default male AI host has the same sound, tone, cadence, and mannerisms that listeners associate with him.

Google disputes the allegations and has publicly maintained that the NotebookLM voice is based on a professional voice actor whom Google hired. At this stage, the court has not determined which side's factual account is correct.

What Is Google NotebookLM?

NotebookLM is a Google AI product that allows users to upload source materials and generate summaries, explanations, and AI-narrated “Audio Overviews.” These Audio Overviews can resemble conversational podcasts featuring synthetic hosts discussing the uploaded materials.

The product has attracted attention because its AI-generated hosts sound conversational and employ humanlike pauses, reactions, filler words, and vocal mannerisms.

Greene contends that NotebookLM's male host does more than sound generally human. He alleges that the host sounds sufficiently like him that ordinary listeners may reasonably believe Greene provided the voice or authorized its use.

That distinction—between a generic synthetic voice and an imitation of an identifiable person—is likely to become central as the litigation proceeds.

What Claims Did Greene Bring Against Google?

Greene filed his complaint in California state court on January 23, 2026. He asserted four state-law causes of action:

1. Statutory right of publicity under California Civil Code Section 3344. Greene alleges that Google knowingly used his voice or likeness for a commercial purpose without consent.

2. Violation of California's Unfair Competition Law under Business and Professions Code Section 17200.Greene alleges that Google failed to disclose that NotebookLM's male voice was engineered to imitate his distinctive voice.

3. Common-law right of publicity. Greene claims that Google commercially exploited a core attribute of his identity—his distinctive and recognizable voice.

4. Unjust enrichment or quasi-contract. Greene alleges that Google received a commercial benefit by exploiting the value of his voice and professional persona without paying him.

Google removed the lawsuit to the Northern District of California on March 2, 2026. Google argued that federal jurisdiction existed because the Copyright Act completely preempted Greene's state-law causes of action.

Google also filed a motion to dismiss. Greene responded by moving to remand the case to California state court.

The Central Legal Issue

The primary issue before the federal court was:

Are Greene's California voice-misappropriation claims really disguised copyright claims that belong in federal court, or are they independent state-law claims protecting his voice, identity, and persona?

The answer depended on whether the alleged injury arose from the copying of copyrighted sound recordings or from the imitation and commercial exploitation of Greene's uncopyrightable voice and identity.

This created an important distinction between AI inputs and AI outputs:

• The alleged inputs may include recordings of Greene's radio and podcast performances that Google supposedly copied or used for AI training.

• The alleged output is the synthetic NotebookLM host that allegedly sounds like Greene and causes listeners to associate the voice with him.

Google focused on the alleged input. Greene focused on the allegedly imitative output.

The court accepted Greene's characterization for purposes of determining federal jurisdiction.

The Rule: Copyright Preemption Under Ninth Circuit Law

The Copyright Act can preempt certain state-law claims. Under the Ninth Circuit's two-part test, a court asks:

1. Does the subject matter of the state-law claim fall within the subject matter of copyright as described in Sections 102 and 103 of the Copyright Act?

2. Are the rights asserted under state law equivalent to the exclusive rights granted to copyright owners under Section 106?

Both requirements ordinarily must be satisfied for Copyright Act preemption.

The subject-matter analysis is broader than determining whether the plaintiff could obtain a copyright registration. Some material may fall within the general subject matter of copyright even if copyright law does not ultimately protect it.

But the court emphasized an important limitation: it is not enough for a state-law claim merely to involve or reference a copyrighted work. The court must identify the actual basis of the alleged misappropriation.

If the claim is based on unauthorized reproduction or distribution of a fixed sound recording, copyright preemption may apply. If the claim instead protects a person's name, likeness, voice, identity, or persona, the claim may remain outside copyright law.

The Court's Analysis

1. Greene's claims concern his voice and identity

The court examined the substance of Greene's four causes of action. Each claim focused primarily on Google's alleged use or exploitation of Greene's:

• Voice;

• Likeness;

• Identity;

• Persona; or

• Distinctive vocal characteristics.

These are personal attributes rather than copyrighted works.

Although Greene's complaint contains allegations that Google may have copied recordings to train or refine NotebookLM, the court concluded that copying was alleged principally as a method by which Google accomplished the voice misappropriation.

The alleged injury was not simply that Google reproduced and distributed one of Greene's podcast episodes. The alleged injury was that Google produced an artificial voice that listeners allegedly identify as Greene.

2. A voice is not copyrightable

The court relied heavily on the Ninth Circuit's landmark decision in Midler v. Ford Motor Co., 849 F.2d 460 (9th Cir. 1988).

In Midler, Ford wanted to use a Bette Midler song in a commercial. After Midler declined to participate, the company hired a sound-alike singer and instructed her to imitate Midler's performance as closely as possible.

The Ninth Circuit held that a distinctive voice is a protectable attribute of identity even though the voice itself is not copyrightable. As that court explained, a voice is highly personal and is not merely a work of authorship.

The Greene court found the principle applicable to artificial intelligence. In a particularly significant passage, Judge Breyer characterized Greene's allegations as being like those in Midler, except that AI allegedly performed the imitation instead of a human sound-alike.

Google argued that Midler should not control because that case involved a human performer. The court rejected that distinction. It reasoned that Midler concerned voice imitation generally and did not establish a rule limited to human imitators.

3. A voice is different from a fixed vocal performance

The court distinguished between a person's general voice and a particular recorded vocal performance.

A specific radio broadcast, podcast episode, song, or recorded performance may be fixed in a tangible medium and protected by copyright. A person's voice as a general attribute of identity is not fixed in that same manner.

This distinction matters because a sound recording can contain an entire particular performance, but no single recording contains the entirety of a person's voice, speaking style, cadence, tone, and identity.

Greene was not limiting his claims to Google's reproduction of one identifiable podcast episode. He alleged that the synthetic host was substantially similar to his voice generally.

That allegation moved the asserted injury away from copyright infringement and toward right-of-publicity law.

4. The claims target AI output rather than merely AI training data

The court's input-output distinction may become one of the most frequently cited aspects of the decision.

AI copyright cases commonly ask whether developers copied protected works when assembling training datasets. Those cases focus on what went into the model and whether the acquisition or use of training material infringed copyrights.

Greene's claims present a different question: What happens when an AI system generates a voice that allegedly imitates an identifiable person?

The court held that Greene's claims, as pleaded, target the product's output. Any alleged copying of recordings was a means of producing that output—not the core injury for which Greene sought state-law relief.

This does not mean AI training practices are legally irrelevant. Evidence about training data, model development, voice actors, prompts, fine-tuning, and internal design decisions could become highly relevant in discovery. It means only that Greene framed his present causes of action around the resulting imitation of his identity.

5. Greene was entitled to frame his complaint under state law

Under the well-pleaded complaint rule, federal jurisdiction ordinarily depends on the claims appearing on the face of the plaintiff's properly pleaded complaint.

The court observed that a plaintiff may selectively plead state-law causes of action to avoid federal jurisdiction. The mere possibility that Greene might also have asserted a copyright claim did not transform his chosen claims into federal claims.

Google invoked the doctrine of complete copyright preemption, under which a nominal state-law claim may be treated as a federal copyright claim from its inception. But complete preemption applies only if the requirements for copyright preemption are satisfied.

Because Google did not establish the first element—copyright subject matter—the court held that complete preemption did not create federal jurisdiction.

6. The court did not decide the equivalency element

The Ninth Circuit's second preemption requirement asks whether the state-law claim seeks to protect rights equivalent to the reproduction, adaptation, distribution, performance, and display rights protected by Section 106.

The court did not reach that issue.

Once it concluded that Greene's claims did not satisfy the subject-matter element, the preemption argument failed. There was no need to decide whether the California claims also contained an “extra element” making them qualitatively different from copyright infringement.

The Court's Conclusion

The federal court granted Greene's motion to remand because Google did not carry its burden of establishing federal subject-matter jurisdiction.

The court held that Greene's claims were based on the alleged output of Google's AI product—an imitation of Greene's voice, likeness, and identity—rather than merely the alleged copying of copyrighted recordings used as inputs.

Because a voice is not copyrightable subject matter, the court found the first part of the Ninth Circuit's copyright-preemption test unsatisfied.

The action will therefore proceed in California state court.

What the Court Did Not Decide

The remand decision should not be mistaken for a final ruling on liability.

The court did not decide:

• That Google copied Greene's recordings;

• That Google trained NotebookLM on Greene's voice;

• That NotebookLM's male host actually imitates Greene;

• That listeners would reasonably identify the synthetic voice as Greene;

• That Google knowingly used Greene's identity;

• That Google received a legally recoverable benefit from the alleged imitation;

• That Greene suffered compensable damages;

• That Greene is entitled to an injunction; or

• That Google violated California's right-of-publicity laws.

The court also did not resolve Google's motion to dismiss. Because the federal court concluded that it lacked jurisdiction, it left the merits of Google's challenges for the state court to address.

Why This Decision Matters for AI Voice-Cloning Cases

Voice rights and copyrights are not the same thing

Copyright may protect a particular recording, but it generally does not grant a person exclusive ownership of the abstract sound of his or her voice.

A distinctive voice may instead receive protection through:

• Statutory right-of-publicity laws;

• Common-law misappropriation doctrines;

• False endorsement or trademark theories;

• Unfair-competition laws;

• Contractual restrictions;

• Privacy laws; and

• Emerging digital-replica legislation.

This means a voice-related dispute can survive outside copyright law even when copyrighted recordings may have played some role in creating the imitation.

AI does not necessarily escape traditional sound-alike law

For decades, courts have addressed advertisements that imitate celebrity voices using human performers. The Greeneorder suggests that replacing a human sound-alike with a synthetic AI voice does not necessarily eliminate the underlying right-of-publicity concern.

The relevant question may be whether the output appropriates an identifiable person's vocal identity—not whether the imitation was produced by a person, algorithm, model, or combination of technologies.

The distinction between copying and imitation will be critical

Future AI disputes may turn on precisely what the plaintiff alleges:

• Copying and distributing a specific sound recording;

• Training a model on copyrighted recordings;

• Cloning a specific recorded performance;

• Generating a sound-alike voice;

• Suggesting a false endorsement;

• Using a voice in advertising;

• Providing users with a tool capable of generating deceptive replicas; or

• Commercially exploiting a person's identity.

These theories may overlap factually but produce different preemption, jurisdiction, liability, and damages analyses.

Pleading strategy can determine the forum

The decision illustrates how the framing of a complaint can affect whether an AI dispute proceeds in state or federal court.

A complaint centered on unauthorized copying of protected recordings may support federal copyright jurisdiction. A complaint centered on the unauthorized commercial use of a voice, identity, or persona may remain in state court.

Plaintiffs must nevertheless plead carefully. The Greene court noted that, to the extent the unfair-competition claim were based solely on copying, that portion could be preempted. Calling a copying claim “unfair competition” does not necessarily prevent preemption.

What Must Greene Prove in California State Court?

Remand gives Greene a state-court forum, but he still faces significant factual and legal burdens.

Identifiability

Greene will likely need to show that NotebookLM's male voice is sufficiently distinctive and similar that listeners identify it with him.

Evidence could include:

• Testimony from listeners and broadcasting professionals;

• Messages from people who independently believed the voice was Greene's;

• Expert acoustic or forensic comparisons;

• Evidence concerning cadence, pitch, phrasing, pauses, mannerisms, and intonation;

• Consumer surveys; and

• Evidence concerning the public recognition of Greene's voice.

A general resemblance to a public-radio broadcaster may not be enough. Greene will need to connect the allegedly imitative characteristics specifically to his identity.

Knowing use

California Civil Code Section 3344 generally requires a knowing use of another person's name, voice, signature, photograph, or likeness for advertising, selling, or soliciting purposes without prior consent.

Google is likely to argue that it hired a professional voice actor and did not knowingly use Greene's voice. Greene may attempt to prove knowledge through internal communications, design documents, voice-selection criteria, testing materials, training records, or evidence that developers expressly referenced him.

Commercial purpose

Greene alleges that the voice was incorporated into a commercially valuable Google product. Google may dispute whether the alleged use satisfies the statutory requirements for advertising or commercial exploitation.

The common-law right-of-publicity claim may not be identical to the statutory claim, making the precise purpose and context of the use particularly important.

Causation and damages

Greene will need to establish the economic or personal injury caused by the alleged misappropriation.

Possible damages theories could include:

• The fair-market value of a voice license;

• Google's profits attributable to the alleged use;

• Lost endorsement or voiceover opportunities;

• Injury to professional identity;

• Harm from being associated with content Greene did not create or approve; and

• Statutory remedies available under California law.

Attribution will be difficult. NotebookLM's value comes from many features, and Google may argue that any resemblance to Greene's voice contributed little or nothing to the product's revenue.

The source of the synthetic voice

Discovery may focus heavily on how Google created the voice. Relevant questions may include:

• Who was the professional voice actor?

• What instructions was the actor given?

• Were Greene's recordings included in any training dataset?

• Were his recordings used during model evaluation or fine-tuning?

• Did developers reference Greene or NPR-style broadcasting?

• Were users or test subjects asked whom the voice resembled?

• Did Google receive complaints or inquiries about the similarity?

• Did Google modify the voice after learning of Greene's objections?

These facts could materially affect knowledge, intent, causation, remedies, and the credibility of both parties' positions.

Google's Likely Defenses

Google may pursue several defenses after remand.

First, it can argue that the synthetic host is based on the voice of a consenting professional actor and was not modeled on Greene.

Second, Google may contend that the asserted similarities—such as cadence, tone, conversational pauses, or a public-radio delivery—are common vocal characteristics rather than attributes uniquely identifying Greene.

Third, Google may challenge whether the voice was used for advertising or endorsement purposes. It could argue that a voice embedded within a product is different from an advertisement falsely representing that Greene endorsed Google.

Fourth, Google may challenge Greene's ability to prove measurable damages or profits attributable to the alleged similarity.

Fifth, Google may renew its preemption arguments against particular allegations or theories if Greene later attempts to recover based directly on the copying of his sound recordings.

Finally, Google may assert constitutional, statutory, or other defenses depending on how the state court interprets the complaint and how the factual record develops.

Where Is the Case Headed Next?

The immediate next step is the return of the case to the California state court where Greene originally filed it. The state court will then address the merits, including arguments raised in Google's federal motion to dismiss if Google renews them under California procedure.

Several possible paths exist.

1. Google may challenge the sufficiency of the complaint

Google will likely argue that Greene has not adequately pleaded one or more elements of his statutory publicity, common-law publicity, unfair-competition, or unjust-enrichment claims.

The state court may dismiss some claims, allow Greene to amend, or permit the case to proceed.

2. The case could move into technically demanding discovery

If Greene survives the pleading stage, discovery may become the most consequential phase of the case.

The parties may seek evidence concerning training datasets, voice-model architecture, development records, voice actors, internal communications, product testing, and Google's knowledge of the alleged resemblance.

Disputes may arise over trade secrets, confidential model information, proprietary datasets, and the appropriate scope of technical discovery.

3. Expert testimony may determine whether the voice is identifiable

This case is unlikely to turn only on whether the voices sound similar to the judge or jury.

Experts may analyze acoustic features and conduct listener-perception studies. The parties could dispute whether current voice-forensic technology can reliably determine whether an AI-generated voice was trained on or derived from a particular person.

Greene may emphasize real-world listener reactions. Google may argue that subjective impressions are insufficient, especially where the voice was created from a hired actor.

4. The parties may seek an early settlement

Voice-misappropriation litigation can present reputational and discovery risks for both sides. A settlement could include:

• Monetary compensation;

• Modification or replacement of the synthetic voice;

• Restrictions on future use;

• A disclaimer;

• No admission of wrongdoing;

• Confidentiality provisions; or

• Mutual releases.

A settlement would avoid a definitive ruling on whether and when an AI-generated sound-alike violates California law.

5. The case could produce influential state-law precedent

If the case proceeds to a substantive ruling, it may help define how California's right-of-publicity laws apply to generative AI voices.

The most important unresolved questions include:

• How similar must an AI voice be before it appropriates identity?

• Must the developer intentionally target a particular person?

• Is hiring a voice actor a complete defense if the final output resembles someone else?

• What evidence establishes that an AI model was trained on a person's recordings?

• Does embedding an imitative voice in a product constitute commercial use?

• What damages are available when the output can generate unlimited new speech?

• Can a court order a developer to modify or discontinue an AI voice?

These questions were not resolved by the remand order, but Greene could become an important vehicle for addressing them.

Key Takeaways for Creators, Voice Actors, and AI Companies

For broadcasters, podcasters, and performers

A person may have enforceable rights in a distinctive voice even when the voice itself cannot be copyrighted.

Professionals who rely on their voices should consider:

• Registering copyrights in important recordings;

• Using written agreements governing AI training and synthetic replicas;

• Monitoring AI products for recognizable imitations;

• Preserving examples of allegedly imitative outputs;

• Collecting unsolicited listener reactions;

• Documenting licensing rates and lost opportunities; and

• Evaluating copyright, publicity, false-endorsement, contract, and unfair-competition theories separately.

For AI developers

Obtaining permission to use training recordings may not eliminate all legal risks. A developer could potentially have lawful access to recordings but still face claims if the resulting output appropriates a recognizable person's identity.

Risk-management measures may include:

• Obtaining express synthetic-voice and digital-replica rights;

• Documenting the source and scope of voice-model consent;

• Testing outputs for resemblance to recognizable individuals;

• Maintaining records of voice development and model provenance;

• Providing an objection or takedown process;

• Avoiding prompts or instructions targeting a particular person's voice;

• Reviewing whether outputs imply endorsement; and

• Updating licenses to address training, cloning, fine-tuning, generated speech, sublicensing, and post-termination use.

For lawyers handling AI voice cases

Counsel should identify the protected interest before choosing claims.

Is the dispute about a copied recording, an imitated performance, a recognizable voice, false endorsement, AI training, breach of a license, or some combination of these?

The answer may determine:

• Whether the case belongs in state or federal court;

• Whether Copyright Act preemption applies;

• What evidence is required;

• What remedies may be available; and

• Whether the plaintiff can obtain injunctive relief.

Final Thoughts

Greene v. Google does not establish that an AI-generated voice violates California law merely because it resembles a real person. It does, however, reject the argument that every AI voice dispute involving recorded training material necessarily becomes a federal copyright case.

The ruling preserves a meaningful role for state right-of-publicity law in the generative AI era. A copyrighted recording and the human identity expressed through that recording are not the same legal interest. Copyright protects fixed creative works; publicity law can protect the commercial value of identity.

The central lesson is straightforward: using copyrighted recordings as AI inputs and generating a recognizable human identity as an AI output may create different causes of action.

As artificial voices become more realistic, courts will increasingly be asked to decide when inspiration, simulation, and general vocal style cross the line into legally actionable appropriation. Greene provides an early answer on one threshold issue: an AI voice-imitation claim can proceed under California law without being swallowed by the Copyright Act.