Strategic Analysis | September 19, 2026

A strange thing is beginning to happen to the economics of fame.

In an August 2026 SEC filing, Agassi Sports Entertainment described an artificial-intelligence platform that is expected to let users interact with AI agents trained on proprietary content from Andre Agassi and longtime coach Darren Cahill. The company said the agents are being programmed to “think, answer and sound” like them. It separately disclosed agreements with an AI company to create AI versions of Agassi and Cahill, alongside 15-year name-and-likeness licenses.

That is more than another celebrity endorsement deal with an AI feature bolted onto it.

A traditional endorsement licenses a person’s identity to sell something. A sufficiently sophisticated digital replica can make the identity itself part of the product.

That distinction is going to rewrite talent contracts.

For decades, lawyers could negotiate “name, image, likeness and voice” clauses primarily around reproduction: Can the brand use the photograph? Can the studio distribute the performance? Can the league put the athlete in an advertisement? Can the record label exploit the recording?

Generative AI introduces a fundamentally different question:

Can the counterparty use what it captured from you to create something you never actually said, did, sang, performed or approved?

That is why “consent” is becoming so valuable. But the word alone is not enough.

The real asset is becoming the architecture of consent: consent to capture, consent to train, consent to generate, consent to alter, consent to distribute, consent to sublicense, consent to retain, and sometimes consent to do it all again in another project.

A one-line likeness grant used to be boilerplate.

In the AI era, it can be a master key.

The easiest way to misunderstand digital-replica rights is to focus entirely on who “owns” the replica.

Ownership matters. But control can matter more.

A producer might own footage of an actor. A brand might own the commercial it shot with an athlete. A game publisher might own recorded performance-capture data. A company might own the AI model into which licensed material was incorporated.

None of those propositions necessarily answers the separate question whether the company is authorized to use those materials to generate new performances by an identifiable person.

That distinction is already visible in collective bargaining.

SAG-AFTRA’s current agreements increasingly separate the creation of a digital replica from its subsequent use. Under the 2025 Commercials Contracts, for example, performer consent is required before a digital replica is created, informed consent is required before it is used, the intended use must be reasonably specifically described, and using the replica to generate a performance triggers compensation. The deal also includes security and eventual destruction requirements for replicas that producers are no longer authorized to retain.

The 2024 Sound Recordings Code similarly requires clear and conspicuous consent in a writing separate from the artist’s underlying performer or royalty agreement, with reasonably specific details about the intended use of a digitally replicated voice.

And the 2026 TV/Theatrical agreement added protections involving digital-replica security, transfer of ownership, dubbing, biometric information and synthetic performers.

There is a pattern here.

The market is moving from ownership of content toward authorization of performance.

That may prove to be one of AI’s most important changes to entertainment economics.

A company can own yesterday’s footage without necessarily owning tomorrow’s performance.

The economics explain why this fight is so intense.

Imagine an athlete spends six hours filming an advertising campaign. Under the old model, the resulting content has finite productive capacity. There are only so many photographs, clips and recorded lines.

With a sufficiently capable digital replica, the same session can become raw material for thousands of future outputs: localized commercials, personalized messages, social videos, interactive avatars, product demonstrations, foreign-language advertisements and content that did not exist when the athlete went home.

For the company, that is extraordinary operating leverage.

For the talent, it creates an entirely different economic bargain.

The license is no longer merely permission to exploit work already performed. It can become an option on future performance without future labor.

That means a digital-replica clause should not automatically be priced like an ordinary footage release. The relevant economic question is not simply what the capture session was worth. It is what future engagements, endorsements, licensing opportunities, approvals and bargaining leverage the replica may replace—or create.

That does not mean every AI use substitutes for employment. Some would never be economical without automation. A digital coach could answer millions of individualized questions that Andre Agassi could never personally answer. An actor could theoretically license a voice replica into markets or languages that would otherwise generate no work at all.

AI can create new inventory, not merely eliminate old inventory.

But that strengthens the case for treating digital-replica rights as their own commercial asset. The worst contract gives away an entirely new line of business as an incidental right buried inside an old one.

The next generation of AI agreements will have to become more precise because “AI use” is not one activity.

At minimum, sophisticated contracts increasingly need to distinguish four verbs:

Capture. Train. Generate. Distribute.

And usually a fifth:

Transfer.

A performer might agree to be scanned but not agree that the scan may train a general-purpose model. An athlete might permit a brand to generate predetermined advertisements but not an interactive chatbot. A musician might authorize a synthetic vocal performance for one track without authorizing a reusable voice model. An actor might authorize a replica for one production without allowing it to migrate with a corporate asset sale into an unrelated business.

Those are economically different permissions.

California law already reflects part of that concern. Labor Code § 927, effective for covered new performances beginning January 1, 2025, can render certain digital-replica provisions unenforceable where a personal-services agreement permits a replica to substitute for work the person otherwise would have performed, lacks a reasonably specific description of the intended uses, and the individual was not represented by qualifying legal counsel or a union whose collective-bargaining agreement addresses digital replicas.

The statute is narrower than it is sometimes described. It does not simply outlaw broad AI licenses. Representation matters. So does contractual specificity. And its express elements matter.

That nuance may be the real strategic significance of the statute: California has not declared that talent cannot sell digital-replica rights. It has made the process by which those rights are sold more important.

The difference between “I consent” and “I consent to this” is becoming legally and economically significant.

Federal law remains unsettled.

There is no comprehensive federal digital-replica statute currently in force. But two important federal developments point in the same direction.

In its Digital Replicas report, the U.S. Copyright Office recommended that Congress create a federal right protecting individuals against unauthorized replicas. More revealingly for contract lawyers, the Office recommended allowing people to license their image and voice rights without permitting them to assign those rights away outright.

Its reasoning went directly to bargaining leverage. The Office warned that unequal bargaining power or knowledge in employment and talent agreements could cause individuals to surrender control of their personas for long periods based on decisions made early in their careers. It suggested relatively short licenses—such as five or ten years outside collective bargaining—followed by an opportunity to renegotiate.

The Senate-reported 2026 NO FAKES Act follows a remarkably similar architecture. The pending bill would create a federal digital-replication right that is nonassignable during an individual’s lifetime but licensable, and its reported text generally contemplates written licenses of no more than ten years containing a reasonably specific description of intended uses, subject to different treatment for collective-bargaining agreements. The Senate Judiciary Committee advanced the bill in June; as of September 2026, it remains legislation rather than enacted law.

That distinction matters. A proposal is not law.

But proposals can still reveal the direction of travel.

The emerging policy instinct is not “nobody may commercialize a digital identity.” It is closer to: a person should be able to commercialize it without permanently selling himself or herself out of the equation.

That is a materially different model from traditional intellectual-property assignment.

Hollywood has an institutional advantage in this transition: collective bargaining.

Sports is more fragmented.

Professional athletes may have union-negotiated group licensing rights in certain contexts, but their endorsements, personal media ventures, appearances, creator businesses and post-career licensing arrangements can involve an entirely different collection of contracts. College athletes add NIL agreements, school and conference rights, collective licensing and sponsor restrictions to the mix.

AI can cut across all of them.

The Agassi Sports Entertainment filings offer an unusually concrete glimpse. The company’s June 30, 2026 Form 10-Q says it entered separate agreements for AI versions of Agassi and Cahill and has long-term name-and-likeness licenses supporting an AI-powered racket-sports platform.

That model creates questions that an ordinary endorsement contract may never have been drafted to answer.

Can the AI replica recommend a product from a sponsor’s competitor? Can it answer questions about gambling, politics, health or another sensitive subject? Can it generate statements the athlete would never personally make? Does approval attach to every response, categories of responses or only the system’s initial parameters? Who bears the risk of hallucinations? Can the replica survive termination of the underlying endorsement? Can a buyer of the company acquire the AI system without acquiring the necessary consent to continue generating outputs?

The familiar NIL vocabulary—name, image and likeness—starts to feel incomplete once the licensed asset can talk back.

For athletes, actors, musicians and creators alike, reputational control becomes economic control because generative systems do not merely reproduce identity. They can exercise it.

There is another reason contracts are about to matter more.

Copyright law and identity rights do different jobs.

A copyright may protect a particular photograph, recording, film or other work. It does not necessarily give a person a general copyright in the sound of his voice or the appearance of her face.

That distinction has already begun producing real litigation.

In Lehrman v. Lovo, two voice actors alleged that an AI voice company used recordings obtained through freelance engagements to create and commercialize clones beyond the uses they had authorized. On a motion to dismiss, the Southern District of New York allowed the plaintiffs’ contract claims, New York right-of-publicity claims and consumer-protection claims to proceed, along with one plaintiff’s copyright claim based on use of her original recording. The court dismissed other theories, including the Lanham Act claims and most copyright theories, while allowing amendment of the training-related copyright allegations.

The procedural posture matters enormously.

The court did not hold after trial that the plaintiffs’ voices were unlawfully cloned. It did not announce a general rule that AI voice imitation violates copyright. It decided which claims had been plausibly pleaded sufficiently to survive a motion to dismiss.

But that is exactly what makes the case useful.

When federal IP doctrines provide an imperfect fit, the mundane question—what did the parties actually agree could be done with the recording?—can become central.

The AI revolution may make contract law look unexpectedly cutting-edge.

There is an important counterargument.

Companies cannot build serious products if every right necessary to operate them can disappear tomorrow.

A studio financing a motion picture needs confidence that lawful distribution will continue after an actor’s digital-replica license expires. A technology company spending millions to build a personalized coaching platform needs some certainty that the central persona will remain available. Advertisers need campaign windows. Game publishers need to support titles after launch.

Consent that can always be revoked instantly is not much of a license.

The Copyright Office recognized exactly this problem. Although it recommended time limits on replica licenses, it also cautioned against using expiration to destroy the continuing exploitation of content lawfully created while the license was valid. In its model, expiration could prevent new uses without necessarily requiring previously authorized works to disappear.

That is probably closer to the durable commercial compromise.

Finished outputs and future generation are not the same thing.

A talent agreement can give a studio durable rights to continue distributing a movie already made without giving that studio a perpetual right to make the actor perform in Movie 7.

A brand can keep an advertisement that was properly approved without retaining an eternal right to manufacture new advertisements.

A game can preserve recorded dialogue without gaining an irrevocable synthetic actor for every sequel.

AI contracting becomes much easier once those concepts are separated.

The video-game industry provides perhaps the clearest illustration of how valuable that distinction can become.

SAG-AFTRA’s 2025 Interactive Media Agreement includes informed-consent requirements for digital-replica uses and, unusually, allows consent to generate new material to be suspended in connection with a strike.

That provision exposes an important second-order issue.

A worker who can strike but whose digital replica can continue performing has considerably less leverage.

In that setting, consent is not merely a privacy right or personality right. It is part of the bargaining machinery of labor itself.

The replica does not get tired. It does not demand overtime. It does not miss a promotional appearance. And, unless the contract says otherwise, it may not respect a picket line.

Digital-replica clauses therefore allocate more than IP rights. They can allocate future bargaining power.

Once digital identities become commercial assets, transactions involving them will create another problem.

Traditional media lawyers obsess over chain of title: Who owns the screenplay? Were the music rights cleared? Who owns the footage?

AI adds a parallel inquiry:

Where is the chain of consent?

A buyer may acquire the model without acquiring the permission to use it.

A company may have authority to generate one class of outputs but not another. A license may terminate on a change of control. Sublicensing might require approval. A replica might have been lawfully created but unlawfully retained. Some outputs might fall within a license while later ones exceed its stated purpose.

SAG-AFTRA’s 2026 TV/Theatrical agreement specifically addresses protection of replica rights when ownership transfers.

Private transactions will increasingly need the same discipline.

Due diligence around AI assets is therefore likely to expand beyond source code, training-data provenance and model ownership. Sophisticated buyers will want the underlying permissions: signed riders, permitted uses, restrictions, term, approval history, sublicensing rights, usage reports, security obligations and deletion requirements.

The expensive question may not be “Do we own this AI?”

It may be “Are we actually allowed to turn it on?”

The best AI clause in the world is less useful if nobody can prove what happened.

Digital-replica disputes are likely to turn on records that traditional talent litigation did not always require: what source material entered a system, what model was used, when an output was generated, which prompt produced it, who approved the output, where it was distributed, whether the use exceeded the authorized description, whether the underlying data remained after termination, and whether copies were actually deleted.

That makes audit rights, usage reports, retention rules and preservation duties more than contractual housekeeping.

They are future evidence.

The 2025 Commercials Contracts already pair consent with security and destruction obligations, while the video-game agreement includes usage reporting designed to help identify noncompliant replica uses.

When an unauthorized synthetic advertisement or performance can reach millions of people before the first demand letter is drafted, speed also matters. Contractual takedown procedures, notice mechanisms and clearly defined restrictions can be worth far more than a beautifully drafted damages clause three years into litigation.

In AI disputes, the first battlefield may be the record of permission.

The most likely evolution is not a universal prohibition on digital replicas.

The commercial upside is too large.

Studios want localization. Advertisers want personalization. Games want dynamic dialogue. Music companies want new forms of exploitation. Sports businesses want interactive coaching and fan experiences. Talent can turn finite personal availability into scalable revenue.

And the agencies themselves are moving toward an authorization model. In August 2026, SAG-AFTRA, CAA, UTA, WME, Cameo and the Music Artist Coalition issued a joint statement emphasizing protection of names, voices and likenesses against unauthorized AI use; Cameo said it does not use talent videos to create or train AI replicas without permission.

So the fight is unlikely to be AI or no AI.

It will be over who controls the meter.

Expect the AI rider to become increasingly important in talent negotiations. Expect separate permissions for replica creation and generation. Expect training rights to be negotiated independently rather than assumed. Expect prohibited-use categories, approval mechanisms, security requirements, deletion obligations and successor provisions to become more detailed. Expect compensation structures to migrate from one-time capture fees toward combinations of minimum guarantees, per-use payments, royalties or revenue participation.

And expect the value of broad old-fashioned grants—“in all media, throughout the universe, in perpetuity”—to be reconsidered when “media” no longer means merely a place to show a performance.

It can mean a machine capable of making new ones.

Generative AI changes the economics of identity because it separates performance from presence.

Once a performer no longer needs to be in the room for a new performance to exist, the critical commercial right is no longer merely the right to record that person once. It is the right to make the digital version act again.

That is why the next generation of talent contracts will not be defined merely by whether they contain the word “consent.”

They will be defined by what the consent actually buys.

AI makes performance scalable. Scalability makes permission scarce.

And in the coming fight over digital replicas, the most valuable right may be the right to say: You can use what I already did. You cannot make me do something new unless we make another deal.