The web’s cards are on the table. Beyond the Meloni–Schlein pact

The agreement between Giorgia Meloni and Elly Schlein not to use artificial intelligence to attribute fabricated words or facts to their opponents is a sign of political civility. It demonstrates that the majority and the opposition can recognise a common rule: public communication must be attributable, verifiable and accountable. But the limitation is clear: a voluntary commitment is binding only on those who choose to abide by it. Those who fabricate information do not sign agreements.

The issue is not limited to politics. Online anonymity has played, and continues to play, a liberal role: it protects dissidents, whistleblowers and those at risk under authoritarian regimes. Yet today, that very same architecture also protects those who mass-produce falsehoods. The task is not to abolish anonymity, but to distinguish between those who must be protected and those who must be identified and prosecuted.

Ordinary citizens are often more vulnerable than leaders. A leader can quickly refute a hoax; an ordinary citizen may be left to face a manipulated audio clip circulating amongst family or at work all on their own. Scams based on voice cloning mainly target the elderly and those less digitally literate: just a few seconds of a voice message are enough to mimic the tone of a family member and ask for money for an emergency. We have even reached the point where we have to agree on a secret word amongst relatives.

Even more serious is the phenomenon of sexual deepfakes, which primarily affects women and girls. Readily available applications transform real photographs into fake sexual images. The investigation by the Rome Public Prosecutor’s Office into an international platform found that the victims included both ordinary people and well-known figures, including Meloni and Schlein. Since 10 October, the new Article 612-quater of the Criminal Code has imposed a prison sentence of between one and five years on anyone who disseminates AI-fabricated images, videos or voices without consent and causes unjust harm. However, a law is only effective if there is an identifiable perpetrator.

There is also a financial cost. In 2024, an employee of Arup in Hong Kong transferred $25 million following a video call in which the finance director and colleagues were digital simulations. It was not the IT systems that had been breached, but human judgement. Deloitte estimates that in the United States, fraud enabled by generative AI could rise from $12.3 billion in 2023 to $40 billion in 2027. Counterfeiting at virtually no cost acts as a tax on trust.

Added to this is the ‘liar’s dividend’: if everything can be fabricated, even what is authentic can be denied. Evidence becomes opinion and doubt turns into an alibi. Digital networks must therefore undergo the same transformation that has already taken place for seas, roads and skies, where flags, registers, number plates and identifiers exist. Online, a distinction must be made between what is recognisable to anyone and what is identifiable by a qualified party.

The public disclosure of a name is never automatic: a person’s real identity should be safeguarded by a third party and disclosed only in cases provided for by law and with adequate safeguards. Alongside identity, the provenance of content is essential: cryptographic credentials and interoperable watermarks document a file’s origin and transformations. They do not certify the truth of a scene, but they do make the history of the content verifiable. The principle is simple: label the synthetic and sign the authentic. This is not about putting the technology on trial. Artificial intelligence is an extraordinary driver of productivity, and accountability and transparency are the steps that bring every new tool to maturity. Trust is an infrastructure: if it is lacking, the value of innovation is diminished and the advantage passes to the counterfeiters.

Europe is moving along two tracks. From 2 August, Article 50 of the AI Act requires, in specified cases, machine-readable markings for generated or manipulated content, and labelling requirements for deepfakes and certain texts of public interest. By the end of 2026, every Member State must offer at least one European digital identity wallet, which is voluntary and user-controlled. On the one hand, tracked content; on the other, certified identities. The link between identity and accountability is missing.

Italy already has many tools at its disposal: the new criminal offence, SPID, the electronic identity card, IT-Wallet and the AGCOM age verification system based on double anonymity, in which the certifying party does not know which service is being accessed and the service receives only the necessary proof.

Three approaches are possible: applying Article 50 without turning it into an indiscriminate obligation; allowing, on a voluntary basis, the use of certified identities or attributes in the most exposed accounts and services, ensuring that accountability can always be traced; and measuring the economic cost of mistrust. The Meloni–Schlein pact provides the political framework, Europe sets the timetable, and Italy already has the infrastructure in place. It must now bring these two tracks together, without exposing citizens’ personal details.