On 5 October 2026, Prime Minister Giorgia Meloni filed an EU trademark application for her own voice. The mark is a four-second recording of her saying "Io sono Giorgia Meloni" twice. Her office confirmed the filing, and wire reports say the stated aim is protection against AI deepfakes. The application was still under examination as of the latest reports. Italian media noted that even an approval would not by itself stop anyone from cloning her voice.
The filing is worth reading as a policy signal. It shows how far a head of government will go to find a legal handle on synthetic media, and it raises a harder question: can intellectual property do the work of an election-integrity rule?
The case for the trademark route
The strongest argument for using trademark law is that it is fast, private and cross-border. A rights holder does not wait for a prosecutor or a regulator. A registered mark gives a takedown claim that platforms already know how to process. For a figure whose voice is a commercial and political identifier, a registration also puts potential misusers on notice. The phrase carries history: it echoes a 2019 rally line and the title of Meloni's 2021 autobiography. That makes the signal clear, even if the legal force is uncertain.
Why it cannot carry the weight
Trademark law regulates use in trade. A mark protects a sign that identifies the commercial origin of goods or services. A deepfake clip that puts false words in a politician's mouth is rarely a use in trade. It is political or deceptive speech, often made by someone with no commercial link to the mark holder. The narrow phrase also matters. A four-second recording covers one utterance, while a voice clone can generate any sentence without reproducing it. A registered recording of one phrase is a weak proxy for a person's voice as a whole.
There is also a free-speech cost if the approach spreads. Political satire and commentary routinely imitate public figures. If a registered voice mark became a tool to suppress parody, the legal tool built to stop fraud would start to chill legitimate speech. Courts would have to weigh that risk case by case, and that is slow when an election is days away.
Italy already has a criminal tool, and it has its own limits
Italy does not rely on trademark alone. Law 132 of 23 September 2025, on artificial intelligence, was published in the Gazzetta Ufficiale on 25 September 2025 and entered into force on 10 October 2025. According to legal commentary, it added article 612-quater to the penal code. The offence covers causing unjust harm to a person by sharing, without their consent, images, videos or voices falsified or altered with AI that can deceive people about their authenticity. The penalty is one to five years in prison. Prosecution generally needs a complaint from the injured party, with exceptions in grave cases.
The steelman for this law is straightforward. Non-consensual synthetic media does real damage, and deterrence needs a penalty with teeth. Meloni's own experience supports that. In 2024 she brought a civil action against a man accused of posting deepfake pornographic images of her, and that case is still pending.
The weakness is structural. A criminal offence works after the fact, against an identifiable person within Italian reach. Election disinformation is typically anonymous, cross-border and fast. By the time a prosecutor opens a case, the clip has done its work. A one-to-five-year sentence also sits close to speech, which is why the "unjust harm" and "capable of deceiving" elements need narrow reading by courts. Satire that no reasonable viewer would take as genuine should fall outside the offence.
The platform layer is where the gap sits
The fastest and most proportionate interventions in an election are made by the services that distribute content. At EU level there are two relevant instruments.
The first is the Commission's election-integrity guidelines under the Digital Services Act, published in 2024. They are addressed to very large online platforms and search engines, which must mitigate risks to electoral processes while safeguarding fundamental rights, including freedom of expression. As law-firm analysis notes, the guidelines recommend clearly labelling AI-generated content such as deepfakes. They are not binding, but a provider that departs from them must show it has taken equally effective measures.
The second is the AI Act. Article 50(4) requires deployers of AI systems that generate or manipulate image, audio or video constituting a deepfake to disclose that the content is artificially generated or manipulated. Artistic, creative, satirical and fictional works get a lighter duty that does not hamper the display of the work.
Labelling is the right default for a pro-speech framework. It adds information instead of removing speech, and it leaves satire and commentary intact. It also works at the speed of a campaign. A label applied when a clip is posted does more than a conviction months later.
What a proportionate response looks like
Three principles follow from the Meloni filing.
- Match the tool to the harm. Use IP for commercial impersonation, such as fake endorsements and cloned voices selling products. Do not stretch it to cover political speech.
- Put speed with the platforms, and keep them accountable. Labelling, provenance signals and reduced amplification of unlabelled synthetic clips during election periods act in hours. Platforms should publish how they apply these measures, so that errors and over-removal can be audited.
- Keep criminal law narrow. Article 612-quater should be applied to deceptive, harmful content shared without consent. Courts should read the genuineness test strictly so that parody and commentary stay protected.
Meloni's filing will tell us something once EUIPO rules on whether a short recording of a spoken phrase can be registered. It does not settle the policy question. The real work on election deepfakes is being done by criminal law after the fact and by platform duties in real time, and the second is where the evidence of performance is thinnest. Regulators should measure whether labelling reaches viewers before the vote, and legislators should resist turning every identity problem into a property right.