Andy Garcia · freelance dev
FRENES
Back to the blog
Legal

Giorgia Meloni Filed Five Seconds of Her Voice as an EU Trade Mark. The Register Shows What It Can and Cannot Stop

October 7, 2026 6 min read

On Monday, October 5, a new entry appeared in the register of the European Union Intellectual Property Office: application 019431219. Applicant: MELONI, GIORGIA, a natural person in Rome. Mark type: sound. Status: application under examination. Reuters reported the filing the same day and said her office had confirmed it.

The mark is an MP3. I downloaded it from the register. It lasts just under five seconds, and a speech-to-text model returns "Io sono Giorgia Meloni, io sono Giorgia Meloni." "Io sono Giorgia" was a line from a 2019 rally, then the title of her 2021 autobiography, as Reuters recalls.

The stated target is deepfakes. In December 2024, Italy's market regulator Consob blocked six websites carrying AI-generated videos of politicians. In one, a fake Meloni promised a "reddito garantito da 40mila euro al mese" for a 250 euro deposit, according to Il Post. Italy has an election due next year. In July, Meloni agreed with Elly Schlein, leader of the opposition Democratic Party, that no party should use AI to attribute invented words to opponents.

I am CTO of CBlindspot, an IP LegalTech that works mostly on patents. The first rule of that field transfers well: a right is exactly as wide as what is written on the register. So I read the register.

She is not the first voice on it

The Italian IP firm that represents Meloni has been filing voices all year. Voice actor Luca Ward's sound mark, "Mi chiamo Luca Ward", was filed on February 12 and registered on May 29. Massimo Corvo's, which covers dubbing, was registered on August 22. Singer Giusy Ferreri's "Sono Giusy Ferreri" went through an opposition period that closed on August 10 and now shows "registration pending".

On Tuesday, October 6, application 019432121 followed: a sound mark in class 41 filed by TAS Rights Management, LLC. On the USPTO record, the same company owns two US applications for the spoken words "Hey, it's Taylor" and "Hey, it's Taylor Swift", filed on April 24 and still awaiting examination. The EU file lasts just over a second; transcribed, it says "Hey, it's Taylor Swift."

So registration is plausible. The EUIPO's guidelines list a "voice saying 'Barca'" among acceptable sound marks.

What the law lets you register

Since October 2017, an EU trade mark no longer needs a graphical representation. Recital 10 of Regulation 2017/1001 allows a sign to be represented "in any appropriate form using generally available technology". Article 4 lists sounds among possible signs, provided they can distinguish goods or services and be represented so that the public can determine "the clear and precise subject matter of the protection".

The EUIPO guidelines in force since July 1, 2026 make that concrete: an MP3 of at most two megabytes, no streaming, no loop. One sentence matters more than the rest: "It is not possible to file a description for sound marks. The representation of the trade mark alone defines the subject matter of the registration."

The mark is the file. Not "her voice". Those five seconds.

What it can stop

Article 9 lets the owner stop third parties "using in the course of trade, in relation to goods or services" an identical sign for identical goods or services, a similar sign where the public is likely to be confused, or a sign that takes unfair advantage of a mark with a reputation in the Union.

For deepfakes, three limits follow.

First, the sign. When the EUIPO compares sound marks that contain speech, its guidelines say it is usually the verbal element that has the greater impact. Their example: two marks with "different voices and intonations" are still similar because both say "Gerivan". On that logic, the protection gravitates around the words "Io sono Giorgia Meloni", even in someone else's voice, more than around her timbre saying anything else. A clone of her voice reading a script about an investment platform does not contain the sign.

Second, the goods and services. The application covers class 9 (downloadable media, electronic publications), class 41 (entertainment, publishing, cultural events) and class 45, where it lists "Organization of political meetings" and "Legal research in the field of public policy". There is no class 36, financial services. Against the kind of scam ad that preceded the filing, the easy case, identical sign for identical services, is not available. She would have to argue similar services and confusion, or a reputation of the mark itself. Neither is easy for a filing from this week.

Third, trade. In Arsenal v Reed (2002), the Court of Justice described use in the course of trade as use "in the context of commercial activity with a view to economic advantage and not as a private matter". A fake clip posted to mislead voters, with nothing for sale, is hard to fit into that sentence. The deepfakes that threaten an election are, mostly, not trade.

Will platforms move faster on a trade mark notice?

That is the usual argument. I read one platform's pages, YouTube's, and they point the other way.

YouTube's trademark page says it "doesn't mediate trademark disputes", will "perform a limited review of reasonable complaints" and remove content "in clear cases of infringement", after forwarding each complaint to the uploader. Its privacy guidelines, by contrast, have a route built for this exact case: "Simulating audio to make it sound like a person is saying something they have not said". Reviewers weigh whether the content is realistic and disclosed, whether it is parody or in the public interest, and whether it shows a public figure "endorsing a product or political candidate".

For a voice deepfake on YouTube, the purpose-built door is the privacy complaint, not the trade mark.

The tool aimed at the generator

The AI Act goes at the problem from the other side. Article 50(4) has applied since August 2: a deployer that uses an AI system to generate or manipulate audio "constituting a deep fake" must disclose it, with a lighter regime for evidently satirical or artistic work. "Deployer" excludes purely personal, non-professional use. Article 50(2) requires providers of systems that generate synthetic audio to mark outputs "in a machine-readable format". The Digital Omnibus, Regulation 2026/1744, gives systems already on the market before August 2 until December 2, 2026 for that marking duty.

A scammer will not label his video. But the marking duty sits on the company whose model produced the audio, and that is where people like me build.

What I would do on Monday if I shipped voice cloning

Check that the voice belongs to the person cloning it, with a recorded consent, not a checkbox. Keep a blocklist of public figures: ElevenLabs does, under the name "no-go voices", a safeguard that "restricts the creation of voice clones that approximate the voices of prominent public figures".

Then add the register to that list. EUIPO's search filters by mark type. On October 7, a search for sound marks returned 598, all years and statuses. Among the 40 most recent, I transcribed at least seven, all filed this year, in which a person says their own name. Each comes with a downloadable MP3, a name and a date. Nothing obliges a voice product to read it. It is a public, dated signal of non-consent, and a speaker-similarity check against a few dozen files costs almost nothing.

Last, never ship a preset, a demo or an ad in which a famous voice says "I am" followed by its owner's name. That is where these marks bite hardest: the same words, in a commercial product, for the goods and services they list.

Meloni's filing will probably work as a signal and a deterrent. As a legal shield against political deepfakes, it is five seconds wide.

Sources

A project like this one?

I design and deploy products like this. Let's talk.

Let's talk