Decentralised News · 2027 planning series · Reviewed 21 November 2026
What Matters
An AI voice or likeness can be technically convincing while its permitted uses remain unclear. A creator’s agreement to one recording, one model or one campaign does not describe every future use. A token representing access or revenue also needs a separate account of the rights it actually conveys.
Before releasing a creator clone, record the person, source material, model version, intended use, distribution routes, payment terms and stop process. These records answer different questions. Combining them into a generic permission to use an “AI twin” makes later review harder.
DN’s original method connects a permission record to a versioned release and a practical stop-request ledger. The local tool counts unresolved actions across declared routes and queued outputs. It does not create a clone, interpret a contract or certify that every copy has been removed.
The Signal: generating a voice and licensing a persona are different tasks
Current voice-cloning tools already distinguish product permissions. ElevenLabs’ Professional Voice Cloning documentation limits creation to the speaker’s own voice and requires verification. It explains a separate sharing route after the owner creates and verifies a clone. Permission from another person does not erase that product’s own creation restrictions.
Its Instant Voice Cloning guidance separately requires the user to confirm rights and consent. Those are product rules, not a universal legal clearance. A vendor checkbox also does not document every recording right, campaign limit or distribution agreement.
The useful business opportunity is controlled production with clear approval and payment. A creator might authorize a bounded narration task or language version. That arrangement needs a reproducible scope and a way to stop further generation without pretending that a payment token contains all the relevant rights.
DN Alpha Thesis
A creator clone becomes commercially useful when permitted uses remain inspectable as the model, operator and distribution change. The important asset is a traceable relationship between permission, approved output and payment, supported by controls that can actually be exercised.
DN proposes reviewing three connected records: the permission boundary, the release version and the stop-request record. Each needs an accountable owner and dated evidence. A favorable result in one record cannot silently resolve a gap in another.
1. Separate the rights before discussing a token
| Layer | What the agreement must address | What does not answer it |
|---|---|---|
| Person and likeness | Whose voice or appearance may be represented, for which uses and term? | Ownership of a sample file |
| Source recording | Who controls the recording and permits its relevant processing? | The speaker appearing in it |
| Model and configuration | Which clone, model version, settings and operators are approved? | A generic vendor account |
| Output and distribution | Which scripts, products, languages, channels and onward uses are permitted? | A technically successful generation |
| Economics | Who receives fees or royalties, for which event and after which deductions? | A token balance or advertised percentage |
| Exit | What must stop, what must be retained and who must confirm action? | A single delete button |
The table is an editorial review method. It is not a declaration that every jurisdiction treats these rights identically. A production can involve several people and agreements: a performer, recording owner, agency, platform, distributor and customer. Record which party can authorize each action instead of assuming that one signature covers the entire chain.
A token can reference a documented licence or payment claim. Its transfer should not be described as transferring ownership of the person. If it purports to give investors future revenue, obtain a separate analysis of that financial arrangement rather than hiding it inside a creative-access feature.
2. Write a use that a reviewer can recognize
“Use my digital twin commercially” leaves important choices open. A more inspectable proposal identifies the product, intended audience, script category, language, distribution channels, term, geography and whether customer interaction is allowed. It also identifies prohibited uses and changes requiring a fresh decision.
For example, permission for a prerecorded tutorial does not itself specify whether the same voice may answer unscripted questions, endorse another product or appear in a political message. A versioned scope lets an operator ask the creator about a real proposed expansion instead of treating silence as an approval.
Keep the creator’s approval evidence in controlled records. The public product can state that the voice is synthetic and explain its intended role without exposing a signature, identity document or private contract. A public hash or identifier may help link records, but it cannot explain the underlying agreement by itself.
Identify the person or organization accountable for the release. Giving a software agent a workflow role does not establish that it can grant rights on the creator’s behalf. The instruction should name what it may prepare, what it may publish and which changes require a separate approval.
3. Distinguish law, recommendations and vendor rules
The US Copyright Office’s July 2024 digital-replicas report analyzes legal gaps and recommends protections and licensing safeguards. Its recommendations are policy analysis, not themselves an enacted nationwide licence. A current transaction needs review under the law that actually applies to its parties and uses.
California provides a jurisdiction-specific example. The Governor’s announcement of AB 2602’s enactment identifies protections addressing digital-replica clauses in performance agreements, including specificity and professional representation. The statute has defined conditions and scope; the announcement should not become a claim that every AI-voice agreement everywhere has the same requirements.
For personal-data processing within GDPR’s scope, the European Commission’s consent guidance distinguishes a specific, informed and freely given consent from a vague permission. It also explains withdrawal for the processing concerned. Consent is one possible lawful basis; the basis for each processing purpose must be identified rather than assumed. A data-processing basis, a likeness licence and a commercial payment contract are related records, not interchangeable labels.
Keep a jurisdiction note in the review: which territories, people and entities are involved, who has assessed the applicable terms, and when that assessment needs refreshing. This article does not determine enforceability or supply a jurisdiction-free contract template.
4. Approve a model version and a release, not just a brand
A product name can remain the same while a model’s output changes. Preserve the model identifier, relevant settings, source-recording version and approved test material. Link the released asset to those records and to the creator’s approval boundary.
ElevenLabs’ current documentation explicitly discusses differences between model versions and recommends comparing outputs with the user’s own content. That supports a practical version-review step; it does not establish that a particular clone is indistinguishable from a person or suitable for every use.
Use tests that match the actual release. A short approved greeting does not establish behavior in an open-ended conversation. A tutorial narration does not test pronunciation of every name or technical term. An agent that chooses its own answers needs separately bounded topics, escalation and output review.
Keep a record of accepted and rejected samples rather than only the best demonstration. Where a creator approves one finished asset, describe that approval narrowly. Do not imply that the creator approved every future output from the model.
5. Make the synthetic role and commercial relationship understandable
A listener should be able to understand what the creator actually contributed and whether the interaction is automated. Design the disclosure around the experience: an audio user may not see a footer, and a clipped video may lose a description. Review the asset and channel that reach the audience.
US FTC influencer guidance explains that material brand relationships should be clear and apparent with the endorsement. A synthetic persona does not remove the need to communicate a relevant commercial relationship. A payment through tokens rather than conventional money does not make that relationship disappear.
The FTC’s consumer-review rule Q&A states that the rule does not impose a blanket ban on AI avatars, while explaining limits around false testimonials and potentially deceptive uses. A creator clone should not claim personal use or experience that never happened merely because the voice sounds familiar.
Keep endorsement approval separate from general voice access. A sponsor’s permission to use the tool does not establish that the creator endorses every generated claim. When the output makes a factual product claim, preserve the supporting evidence and the person responsible for checking it.
6. Price the permitted event and reconcile the recipient
Define what creates a payment obligation: an approved asset, a period of permitted access, a generated minute, a completed campaign or another specified event. Then identify the reporting source and the recipient. An unreviewed generation should not silently become a paid authorized publication.
A royalty percentage needs a base. Gross customer spend, platform receipts, net revenue after refunds and distributable cash can all differ. Identify deductions, reserves, currency, settlement dates and access to supporting records. DN’s separate creator-licensing settlement guide handles that financial waterfall in detail.
A clone agreement adds an operational question: which version and permitted use produced the reported event? Linking a payment to a model or asset identifier can help investigate it, but it does not automatically prove that the underlying use was permitted. Retain the approval link and the actual payment evidence.
Do not use an automatic penalty or token burn as the whole remedy for an unauthorized output. The production team still needs a way to stop dissemination, address affected parties and reconcile any contractual obligation. Money movement alone does not undo a misleading endorsement or restore control of copied media.
7. Plan withdrawal and stopping as distinct actions
A stop request can concern new generation, a specific campaign, an account credential, an active distribution route, queued outputs or retained source material. Identify the actual request and the relevant authority before changing a production system. A familiar synthetic voice should not serve as the sole authentication of that request.
The European Data Protection Board’s consent summary explains the right to withdraw consent and the requirement that withdrawal be as easy as giving it. The legal consequence concerns the processing based on that consent. It should not be reduced to changing an onchain status flag while the affected processing continues elsewhere.
Erasure has its own scope. GDPR Article 17 specifies grounds and exceptions rather than a universal promise that every copy vanishes. Records, backups, previously distributed assets and legally required retention need a documented review under the applicable arrangement.
Operationally, assign each requested action to an owner and obtain evidence of completion. Removing an API key may stop one generator but leave a distributor account active. Disabling future generation does not recall an existing file. Preserve those differences in the response to the creator.
8. Worked synthetic case: one stop request, several incomplete actions
Consider an invented creator who licensed a synthetic voice for approved product tutorials. The permission excludes unscripted endorsements. The team inventories eight controllable generation or distribution routes and fifteen outputs waiting in production queues.
After a documented stop request, six routes have confirmed suspension and ten queued outputs have confirmed quarantine. Two routes and five queued outputs remain unresolved. The DN record shows those counts without turning them into a percentage called “safe” or “compliant.”
The team also identifies forty previously published tutorial files. Those files are outside the queue count and need a separate distribution and rights review. Some may remain permitted under the applicable agreement; others may require action. The calculator cannot determine that legal result from the number forty.
Suppose a contractor says the remaining two routes are inactive, but supplies no current observation. Keep their status unresolved until the responsible owner records the required evidence. A verbal assurance and a verified action are different kinds of record.
Finally, reconcile fees for the already approved tutorials separately from the stop work. A pending payment does not by itself authorize new generation, and a stopped generator does not by itself explain an unpaid approved invoice. These are synthetic decisions, not findings about a real creator or platform.
DN Clone Stop-Request Record
Use anonymous counts for one documented scope and observation. Declare whether the scope, inventory, downstream notices and retention review are resolved. Count only actions with confirmation. No real names, recordings, signatures, contracts or account identifiers are needed.
DN Clone Stop-Request Record
Change the illustrative assumptions below. This worksheet runs locally in your browser. It records your declarations; it does not verify evidence, connect to accounts, submit instructions or grant approval.
The initial numbers are synthetic. Unknown evidence remains unresolved. Downloads contain the values you enter; use aliases and avoid private or identifying information.
9. Keep automation subordinate to the permission record
An assistant can assemble approved scripts, compare versions, flag missing approvals, organize a distribution inventory and draft stop notices for an authorized operator. Those tasks can make the record easier to inspect. The assistant should preserve the original evidence rather than replace it with a confident paraphrase.
Keep access to source recordings, generation systems and publishing accounts separately controlled. A service preparing scripts does not necessarily need the ability to export the underlying voice model or change revenue recipients. A community vote or token balance should not silently expand powers granted by the creator’s agreement.
Review a new channel or model as a proposed change. Record the reason, expected benefit, additional data flow and approval. If the change cannot be matched to the permission boundary, resolve that gap before release. A successful technical integration is not the same event as an approved commercial use.
Methodology and limitations
DN reviewed the linked official product, regulator and government materials on 21 November 2026 and compared this scope with its creator-licensing and provenance articles. The contribution is a permission/version/stop workflow and a local backlog calculator. No clone, creator contract, biometric verification or takedown process was tested.
The calculator subtracts confirmed suspended routes from declared in-scope routes and confirmed quarantined outputs from the declared queue. It flags unresolved evidence states and rejects impossible counts. It does not discover unknown routes, verify a stop request, interpret a retention duty, send notices or prove deletion.
The numerical example is entirely synthetic. Counts are operational quantities, not a risk score. Legal requirements vary by jurisdiction and use, and vendor rules can change. The practical deliverable for a real production is a scoped record showing what was approved, which version was released, what was paid and what remains to be done when permission changes.
Sources and evidence register
Primary sources reviewed on 21 November 2026. The article distinguishes documented facts, DN analysis and synthetic examples. Recheck changeable rules and configurations before acting.
- ElevenLabs: Professional Voice Cloning — Own-voice verification and sharing boundary; model-version differences. Not general legal clearance.
- ElevenLabs: Instant Voice Cloning — Product-specific rights and consent confirmation.
- US Copyright Office: Digital Replicas report — July 2024 analysis and recommendations, not an enacted nationwide licence.
- California Governor: AB 2602 enactment announcement — Jurisdiction-specific context for digital-replica contract protection; statute-specific conditions need legal review.
- European Commission: Legal grounds for processing data — Lawful processing bases; specific, informed, freely given consent and withdrawal where processing relies on consent within GDPR scope.
- FTC: Disclosures 101 — Material commercial relationships and understandable endorsement disclosures in the US context.
- FTC: Consumer reviews and testimonials rule Q&A — AI avatars are not universally banned; false testimonials and deceptive use remain separate questions.
- EDPB: Consent summary — Withdrawal right and ease of withdrawing processing consent.
- EUR-Lex: GDPR — Article 17 erasure grounds and exceptions; no universal promise that all distributed media can be recalled.
Frequently asked questions
Does owning a creator-clone token mean owning the creator's likeness?
The token needs a documented explanation of the access, payment or other rights it conveys. Its ownership alone does not establish a complete likeness licence or ownership of the person.
Can an agency create someone else's Professional Voice Clone on ElevenLabs with consent?
The reviewed Professional Voice Cloning rules require people to create and verify their own voice. They can then use the platform's documented sharing route. Vendor rules and applicable legal permissions both matter.
Does approval of a finished clip authorize all future clone outputs?
Do not assume so. Record whether approval applies to that asset, a particular model and use, or a broader defined production scope. New uses must be reviewed against the actual permission.
Does deleting a clone remove every published copy?
No such result is established here. Generation systems, queued jobs, distributor accounts and existing files need separate action and rights review.
Can a synthetic influencer make up a personal product experience?
A familiar synthetic voice does not turn an invented experience into a truthful testimonial. The cited FTC guidance distinguishes avatar use from false or deceptive endorsements.
Does the stop-request calculator establish legal compliance?
No. It reports declared outstanding actions and unresolved evidence. It does not verify the inventory, interpret contracts, authenticate a person, send notices or prove erasure.




