A voice session used to have an intuitive endpoint. A performer read the approved copy, the recording was edited, and the work went to its named audience. Synthetic speech changes the unit of work. A recording can now become a system that produces new words, new moods, new languages, or new versions long after the room has emptied.
That possibility is not automatically exploitative. A performer may want a licensing business, an accessible version, or a way to correct a line without another studio day. The question is whether the person agreed to that second life, and whether the agreement still means what the parties think it means when use expands.
This is an editorial framework, not a cloning guide or personalized legal advice. It separates performance, contract, platform, and legal questions. A checkbox saying “AI use permitted” cannot carry all four.
A voice performance is more than a sound file
The expressive work in a voice is not only the waveform. It includes words, timing, emphasis, accent, character, emotional temperature, and relationship. A synthetic system can use vocal characteristics to say something the speaker never said. ElevenLabs describes a clone as a representation rather than the original recording that can generate new words. This is a company explanation, not independent proof of quality or safety. Its documentation also says verification does not settle every question about authorized use.
That distinction matters to the performer. “Record this 30-second launch film” requests a defined performance. “Let us make a voice model” involves future performances. “Use the model in paid ads” adds distribution and commercial context. “Keep it for later” adds a retention decision. Related transactions are not the same promise.
The four permissions hidden inside one checkbox
A useful review unpacks the checkbox into four decisions. This is a proposed artifact, not a claim that one form suffices everywhere.
-
Creation. May the producer create and retain a model or other representation from the performer’s voice? The answer should identify the source material and the purpose. A session for one ad does not silently become permission to build a general-purpose asset.
-
Generation. May the representation produce new speech for the named project? Specify the script, language, delivery, character, and whether the performer will review the generated performance. A new sentence or emotional direction can be a new expressive decision even when the timbre sounds familiar.
-
Reuse. May the same asset appear in another cut, campaign, territory, platform, product, or time period? Reuse can change the audience, the commercial value, and the meaning of the performance. It should be a fresh question when the original scope no longer describes the proposed use.
-
Retention and exit. Who may access the representation, for how long, and what happens when the license ends, the project changes hands, or the performer says no to a new use? A deletion promise should identify the copy, the custodian, and the event that starts the clock. “We will not use it” is not the same as “we will no longer keep it.”
This reveals why a yes-or-no field is weak. It records a decision without its object and asks a performer to anticipate future combinations. Informed agreement makes the variables legible before collection.
What the public record actually shows
A union contract is a negotiated position. SAG-AFTRA’s 2025 commercials guidance requires consent before creation and informed consent with a reasonably specific description before use. It ties replica performance to compensation, addresses a maximum period of use, and says retention after the last commercial requires consent or destruction. These are terms of one collective agreement, not every voice job. SAG-AFTRA’s 2025 Commercials Contracts
SAG-AFTRA announced on June 4, 2026 that members ratified the 2026 TV/Theatrical Agreement, effective July 1, 2026 through June 30, 2030. Its current contract page describes protections around replicas, biometric data, security, transfer, and synthetic performers. These are terms of a particular collective agreement, not every voice job or production. SAG-AFTRA’s ratification announcement and 2026 TV/Theatrical Contracts summary
A platform rule is a company position. ElevenLabs’ current documentation describes voice verification and its limitations. Verification can help check identity; it is not a commercial license, reuse fee, ownership decision, or guarantee that downstream users received the same permission. The relationship still needs its own record.
A government report is not the same thing as a statute. The U.S. Copyright Office’s July 2024 report describes the problem and recommends that Congress create a federal law against knowing distribution of unauthorized digital replicas. It also discusses uneven state protections and protected expression. The report is policy advice, not a universal federal permission rule. U.S. Copyright Office, Part 1: Digital Replicas
The inference is modest but important: contract language, platform controls, and legal protections answer different questions. Obligations depend on the agreement, people, location, use, and current law. For a dispute or high-stakes use, consult counsel or the relevant representative.
Build a voice-use card before the session
Before recording, create a one-page voice-use card attached to the brief. It should be readable by the performer, producer, editor, and later-use approver.
| Card field | Record | Stop condition |
|---|---|---|
| Performance | Person, script, language, character, delivery, and whether new speech may be generated | The requested output adds a new character, claim, language, or emotional direction not described |
| Scope | Named project, channels, territories, audience, paid or organic placement, term, and exclusivity | A cutdown, market, platform, or campaign is outside the stated scope |
| Control | Custodian, access list, subcontractors, transfer rules, review rights, and correction contact | The team cannot identify who can generate or approve speech |
| Money and exit | Creation fee, generation or reuse fee, renewal, credit, retention period, deletion trigger, and dispute route | The asset will be kept or reused after the period or payment described |
The point is not paperwork for its own sake. The card gives a reviewer something concrete to compare with a new request. For a Spanish version six months later or a campaign sale, it exposes language, term, audience, compensation, and transfer. A performer’s question about what remains should not depend on memory.
Keep a usage log with the approved version, date, channel, generating party, reviewer, and expiry check. It is not proof of authorization; it exposes mismatch. A blank field should read “unknown” and pause release, not invite inference from silence.
The edge case is where consent earns its meaning
Consider a hypothetical campaign. A brand hires Maya, a voice performer, for a human-read 30-second audio ad. The agreement says “AI use permitted” but names no script beyond the recording day, no markets, no period, and no retention rule. Six months later, an agency requests a Spanish synthetic version announcing a larger product size, for paid placements in a new market.
The version changes words, language, context, audience, and date. Mark it out of scope, return it to the owner and performer, and hold release until the new use and commercial terms are documented. Separately verify the new product claim: fresh consent cannot make an unsupported claim true. The example does not decide legal rights; it shows why a vague field is a poor handoff.
Consent should survive the first “yes”
The humane standard is not “never use synthetic voices.” It is “do not turn a person’s first yes into an unlimited yes by omission.” A performer can choose a narrow, broad, renewable, or no-synthetic-use license. The work is more trustworthy when that choice is visible and revisable.
For producers, the benefit is clarity: fewer surprises, cleaner renewals, and a record distinguishing paid work from an unpriced reserve asset. For performers, it is agency over what their voice can carry. For audiences, it is an honest claim: named scope, custodian, and correction route.
The checkbox can still exist. It just belongs at the end of a sentence that says what, where, when, how, and whose decision it records.
Sources and limitations
- ElevenLabs, “Voice cloning: how it works” — checked September 2, 2026. Supports the company’s description of a cloned voice as a representation that can generate new speech and its explanation of verification. Limitation: vendor documentation and product policy, not independent testing, a contract, or a legal conclusion.
- SAG-AFTRA, “2025 Commercials Contracts” — checked September 2, 2026. Supports the contract-specific consent, specificity, compensation, maximum-use-period, security, and destruction terms described above. Limitation: applies to the stated collective agreement and covered work, not every performer or production.
- SAG-AFTRA, “2026 TV/Theatrical Contracts” — checked September 2, 2026. Supports the current landing-page summary of replica, biometric-data, security, transfer, and synthetic-performer provisions. Limitation: applies to the collective agreement and covered work, not every voice job or independent legal outcome.
- SAG-AFTRA, “SAG-AFTRA Members Approve 2026 TV/Theatrical Contracts Tentative Agreement” — checked September 2, 2026. Supports the June 4, 2026 ratification announcement and the July 1, 2026 effective date through June 30, 2030. Limitation: union announcement and contract-specific status, not a universal rule or personalized legal advice.
- U.S. Copyright Office, “Copyright and Artificial Intelligence, Part 1: Digital Replicas” — checked September 2, 2026. Supports the report’s description of digital-replica legal issues, existing uneven protections, and recommendation for federal legislation. Limitation: July 2024 policy report, not personalized legal advice or a statement that a proposed federal law is in force.