AI Dubbing: who owns the voice?

What is AI dubbing?

AI dubbing refers to AI-supported processes in which existing voice recordings are processed as input in order to create new language versions, synthetically generated voices or variations. Put simply: the recording is not merely edited or mixed. It may be fed into an AI process that derives further versions from the same performance — up to and including the use of a voice replica for automated outputs.

Why AI dubbing is legally sensitive right now

The current debate is being shaped by the publicly discussed dispute between Netflix and German dubbing actors. The case shows what is at stake in practice: the scope and specificity of contracts and consents — in particular, whether voice recordings may be used not only for work-specific dubbing, but also as input for AI processes.

This is precisely where the legal sensitivity lies. A person who originally released their voice for a traditional dubbing performance has not automatically consented to every subsequent AI use. This is especially true where new synthetic outputs are created from the recording that go beyond the original production.

A person’s voice is protected by general personality rights, and voice recordings are also regularly personal data. The decisive question is therefore not what is technically possible, but whether the consent and the chain of rights actually cover the specific AI use.

The following levels are particularly important:

• the original recording,
• use as input,
• synthetic or otherwise derived outputs,
• further use and commercial exploitation.

Why standard clauses are often not enough

Many dubbing and production contracts are tailored to traditional post-production steps. This is often insufficient for AI dubbing. Even the use of a voice recording as input in AI processes may constitute an independent type of use requiring consent. The derived results and their exploitation are also decisive.

In practice, this means asking:

• Scope: Does the consent expressly cover AI input use and outputs?
• Control: Are there protective clauses for new statements or uses outside the original context?
• Exploitation: Are outputs, media, duration, transferability and sublicensing clearly regulated?

The key distinction: input and output

A practical guideline for contract review is to distinguish between input and output. This helps divide AI clauses into two questions that can be assessed separately.

First: may the voice recording be used as input in AI processes at all?

Second: may the outputs generated from it be exploited, and if so, to what extent?

Many contractual provisions are drafted with a specific work in mind and say little about whether derived results may later be reused, scaled or transferred into new contexts.

Data protection must be reviewed separately

In addition to personality rights, data protection law must be assessed as a separate issue. A contractual transfer of rights or consent for the work does not replace a legal basis for processing under the GDPR.

Voice recordings are regularly personal data. This raises the usual GDPR questions:

• who processes the data and in which role,
• on which legal basis,
• for which purpose,
• for how long the data are stored, and
• whether disclosures or later changes of purpose, particularly for AI training or reuse, are covered.

For companies, this means:

• contractual clearance and data protection clearance are not the same thing,
• processing must be based on a clearly defined purpose,
• retention periods, disclosures and changes of purpose must be considered at an early stage.

Protective clauses often make the difference

Protective clauses become particularly important when artificial intelligence is used because artificially generated voice versions can shift a voice from its originally controlled use into new statements and new contexts. Without clear limits, the contractual permission may effectively become unlimited. The risk of shifts in meaning and context lies in reputational damage, misleading attribution or the assignment of a statement to a person who did not make it themselves.

Useful protective mechanisms include:

• prohibitions on use in certain contexts,
• approval processes for new usage scenarios,
• special rules for synthetic or digitally replicated voices,
• separate approvals for advertising, sensitive topics and secondary uses outside the original project.

The greatest risk often arises at the output stage

The commercial use of synthetically replicated voices is particularly risky where the output is attributed to a specific person. In such cases, the issue is not only the unlawful exploitation of the original recording, but also the use of a personality right-protected characteristic with its own economic value.

The Regional Court of Berlin II has clarified that an interference may also exist where the original voice is not used, but rather an imitation generated by artificial intelligence. It also held that the principles governing commercial exploitation in cases of unauthorised image use can be transferred to the voice.

The typical legal consequences may include:

• claims for injunctive relief and removal,
• claims for information and accounting to clarify the scope of use,
• payment claims based on licence analogy or damages,
• disputes over the scope and reach of contractual rights and their economic compensation.

Remuneration and governance must also be considered

AI dubbing changes not only the type of use, but also its economic value. If a single recording performance gives rise to numerous further uses, the question arises whether the original remuneration still reflects the actual exploitation.

Internal governance is equally important. Even well-drafted contracts are of limited help if it remains unclear within the company who approves AI uses, reviews outputs, authorises reuse and documents actual use.

Especially in the case of international language versions and external technology partners, clear responsibilities, defined review paths and traceable documentation are therefore essential.

From August 2026, another compliance layer will be added

For AI-generated or AI-manipulated audio content, a transparency obligation under the AI Act may become relevant from 2 August 2026 – particularly where the content constitutes a deepfake.

This means that AI dubbing will also become a transparency and labelling issue. Companies should therefore not treat the topic in isolation, but as an interplay between contractual design, personality rights, data protection, approval processes and transparency obligations.

Conclusion: AI dubbing needs clear rules before it scales

The real challenge in AI dubbing does not lie solely in the technology, but in establishing a clean rights and approval structure. Traditional contract templates often do not reliably cover AI-specific uses of voices – especially when it comes to input use, synthetic outputs, further exploitation and remuneration.

The key takeaway is therefore clear: anyone seeking to use AI dubbing in a legally robust way should expressly regulate AI-specific releases, formulate protective clauses precisely, take the data protection purpose limitation into account and establish reliable governance processes.

The practical conclusion is equally clear: companies should review existing contract templates, release processes and internal responsibilities at an early stage, before AI dubbing is used on a larger scale or rolled out internationally.