Sound mark and voice deepfakes: how to protect your voice from AI cloning

Why a cloned voice is hard to protect

Your voice is protected in pieces, by rules written for other things, and each piece is easier to enforce when you can prove what was published.

In practice you combine trademark law, personality rights, the GDPR and platform notices under Article 16 of the EU Digital Services Act. Each works better when you can show the audio, where it appeared and when.

A voice is neither a work nor a registered sign

Copyright protects works and, through related rights, performances and recordings; the timbre of your voice is, as a rule, not a work. Trademark law mainly protects registered signs, and few people have registered their voice. That leaves personality rights, which differ by country. Denmark presented a bill on October 8, 2026 to protect faces and voices through copyright from January 1, 2027, if Parliament approves it (Euronews).

How cloned voices are misused: fake endorsements, scams and damaging content

In a fake endorsement, a well-known host seems to recommend a supplement or an investment. In fraud, a cloned voice asks a relative or an employee for money, and our analysis of voice cloning fraud against CFOs covers the corporate version. Then there is content made to mislead. Before the New Hampshire primary of January 21, 2024, robocalls with a deepfake voice of Joe Biden reached voters, and the FCC later imposed a $6 million fine for those deceptive automated calls, not for cloning the voice as such (FCC 24-104).

Registering your voice as a sound mark

If the application succeeds, a sound mark turns one recording of your voice into a registered title, with a number and a date you can cite in removal requests.

What a sound mark is after the EU reform

Article 4 of the EU Trade Mark Regulation (EU) 2017/1001 lets any sign, “sounds” included, become an EU trademark if it distinguishes goods or services and the register can show “the clear and precise subject matter of the protection.” Under Article 3(3)(g) of Implementing Regulation (EU) 2018/626, you file “an audio file reproducing the sound” or an accurate musical notation.

If you record that clip live in the TrueScreen App, the report documents its SHA-256 hash and the date of certification, so a version of your voice is on record before any clone appears. That date does not replace the filing, and it does not prove whose voice it is.

What has been filed: Meloni’s application and earlier voice filings

Giorgia Meloni’s EUIPO application of October 5, 2026 covers a recording in which she says “Io sono Giorgia Meloni” twice, for downloadable multimedia content, cultural and entertainment events, and political events (Il Post; Euronews). It is an application under examination, not a registered mark. According to Il Post, Italian voice actor Luca Ward and singer Giusy Ferreri had already filed applications for sound marks of their voices.

How to file a sound mark

Choose the EUIPO for one title across the EU, or a national IP office if one country is enough. Upload a clean audio file and choose the classes of goods and services with a trademark attorney, since the mark only works inside them and fees depend on how many you list. The office examines the application and publishes it for opposition, so registration is not guaranteed. A registered EU trademark lasts ten years from the filing date and is renewable.

When a sound mark helps with platform takedowns

Article 16 of the Digital Services Act (Regulation (EU) 2022/2065) requires hosting services to run a notice-and-action mechanism and to decide on notices “in a timely, diligent, non-arbitrary and objective manner.” A notice needs a reasoned explanation, the exact URL, your name and email and a statement of good faith. A registered mark can support it only if the content falls within the mark’s scope; otherwise the notice must rest on other grounds, such as personality rights, personal data or national law. Removal remains the platform’s decision, and Article 22 gives priority only to designated trusted flaggers, a status private individuals do not have.

What a sound mark cannot stop

A sound mark protects the recording you filed, for the goods and services you listed, not your voice: a clone saying new words outside commercial use mostly escapes it.

Article 9(2) of the EU Trade Mark Regulation lets the owner stop identical or similar signs used “in the course of trade” for goods or services. Many voice deepfakes meet neither condition: they say new words, and nobody sells anything with them.

No monopoly on a voice

Your timbre, accent and way of speaking stay unregistered, so impressionists, people who happen to sound like you and clones that never play the registered clip are outside the direct scope of the mark, although other rights may still apply.

Clones that never say the registered phrase

A clone says whatever its author types, rarely your four-second name clip. Article 9(2) does reach similar signs that create a likelihood of confusion, and protects a mark that “has a reputation” more widely. Legal experts quoted by Il Post note, however, that a fake clip in which the voice says something else does not infringe the mark. As of October 2026, no court appears to have ruled on a personal sound mark against an AI clone.

Trademark law reaches only commercial use

The same experts point out that a parody would fall under the mark only if used commercially, in an advertisement for example. By the same logic, satire, political commentary and personal attacks stay outside trademark law when nobody uses them commercially, and many voice deepfakes are made with no commercial purpose at all.

Other legal protections against voice deepfakes

Personality rights, data protection, disclosure rules and contracts cover what the trademark misses, and each is easier to use with evidence of the content.

Contracts are the simplest layer: if you lend your voice to studios or brands, have a lawyer draft the clauses on how recordings may be used and whether they may train AI models. A contract binds only the people who signed it.

Deepfake-specific rules: where they exist

In the United States, Tennessee’s ELVIS Act, in force since July 1, 2024, treats the voice as a personal right and addresses AI voice cloning (Holland & Knight). The NO FAKES Act of 2026 (S. 4591) has passed a Senate committee but is not law. In the EU, Italy has already made it a criminal offense to spread AI-falsified voices that cause harm.

Personality rights and the GDPR: your voice as personal data

Your voice is personal data whenever it identifies you or makes you identifiable, under Article 4(1) of the GDPR (Regulation (EU) 2016/679). It counts as biometric data only after “specific technical processing” for unique identification (Article 4(14)), and Article 9(1) then prohibits that processing unless an exception applies. Article 17 lets you request erasure when one of its grounds applies, with an exception for freedom of expression. Whether the GDPR reaches a given publisher, such as a private individual or someone outside the EU, is a question for a lawyer.

Personality rights can protect the voice too: on August 20, 2025, a Berlin court (LG Berlin II, case 2 O 202/24) held that an AI clone of German voice actor Manfred Lehmann’s voice, used in videos, violated his right to his own voice, and awarded €4,000 as a notional license fee (beck-aktuell).

The AI Act: providers mark synthetic audio, professional deployers disclose deep fakes

The EU AI Act (Regulation (EU) 2024/1689) defines a deep fake in Article 3(60) as AI content resembling real people or events that “would falsely appear to a person to be authentic or truthful.” Under Article 50, paragraph 2, providers of systems that generate synthetic audio must mark the output in a machine-readable format. Under paragraph 4, deployers using the system in a professional capacity, not for purely personal activity, must disclose deep fakes, with lighter duties for evidently artistic, satirical or fictional works. Both apply from August 2, 2026, although the Digital Omnibus Regulation (EU) 2026/1744 moves the marking duty to December 2, 2026 for systems already on the market. The exact scope in your case is for a lawyer to assess.

Evidence decides whether any protection works

Every remedy against a voice deepfake gets stronger with the same file: certified proof of the deepfake as it was published, and of your real voice.

A takedown notice, a warning letter, an erasure request and a lawsuit all go better when you can show what was published, where, when and in what context. Once the deepfake is deleted, a clip nobody captured is your word against theirs.

Why screenshots and links are not enough

A screenshot of an audio deepfake has no audio in it, only a frame and a caption, and a link dies as soon as the post comes down. Both are easy to dispute, because a screenshot carries no reliable time and nothing to show it wasn’t edited, as our guide to capturing social media evidence for court explains.

What the evidence must show

Useful evidence of a voice deepfake shows the exact audio or video, the address where it was published, the moment it was captured, and the context (account, caption, comments, view count). TrueScreen captures them together in one forensic acquisition before removal. The certification proves that the content was acquired with forensic methodology at that address and at that moment and that any later alteration would be detectable; it does not prove that the content is true or who made it.

Certify your original voice as a reference

To say whether a clip is a clone, a forensic audio expert works better with authentic recordings of known date and origin. With the TrueScreen App you can record your voice live from the microphone, and the report documents the SHA-256 hash of the recording and the date of certification. Existing files, such as a podcast, imported into the App are sealed with a certified date from that moment, and any later alteration would be detectable; this is not a forensic acquisition and does not certify when or by whom they were recorded. Comparing voices is the forensic audio expert’s job; TrueScreen certifies the acquisition (see our pieces on voice notes as court evidence and source certification versus detection).

Feature

Capture, analyse and certify any online content.

With the TrueScreen Forensic Browser, the page carrying a disputed voice is captured with forensic methodology, audio and video included.

Discover more →

How do you collect certified evidence of a voice deepfake?

TrueScreen certifies the page that carries a voice deepfake while it is still online, and certifies recordings of your real voice made live in the App, so the expert has a reference to compare it with.

The strongest route is the Forensic Browser in Full recording mode, which records audio and video of everything that happens during the browsing session, together with the network log (HAR). TrueScreen, the Data Authenticity Platform, captures the whole session, and the package is sealed with a qualified electronic seal and a qualified timestamp: certified evidence that can be presented in court, with admissibility and weight left to the judge. It runs on macOS and Windows and starts each acquisition with browsing data cleared.

If the clip could disappear quickly, use what you already have installed: the Browser Extension for Chrome and Edge records the tab as video with its audio, and its report lists the URL, the UTC date and a SHA-256 hash. If you expect the other side to contest the capture, use the Forensic Browser.

For audio or video found on a page during a Forensic Browser session, the AI analysis in the Media Inspector (version 2.9.3 and later) estimates how likely it is that the media was generated or manipulated. It is probabilistic and can be wrong, so a person makes the final call, and it does not replace certifying a page that is still online. Anyone can check the seal with the EU DSS tool or on the certification verification page.

One possible scenario: a voice actor finds a supplement ad in which his cloned voice “recommends” the product. His agent captures it in Full recording, with the account and comments. His lawyer files the notice with the sound mark’s registration number, if registered, and the certified package, and an audio expert compares the clip with recordings made live in the TrueScreen App months earlier, so both have documented digital provenance.

1. 1

   Certify the content before you report it

   Capture the page, the audio or video, the account and the comments before you report, comment or contact the author: any of those moves can trigger removal.

2. 2

   Request removal and send a cease-and-desist letter

   File the platform notice under DSA Article 16 with the exact URL, a reasoned explanation, your name and email and a statement of good faith. Cite a registered sound mark only if the content falls within its scope. Your lawyer can write to the author and send a GDPR erasure request.

3. 3

   Keep the evidence file for legal action

   Store the certified package (report, hashes, seal and timestamp) with your certified original recordings. Because it contains third-party accounts and comments, hand it to your lawyer and do not republish it.

Conclusion