A rare cross-party opening
On 24 September 2026, Il Foglio reported that Italy's parties are converging on action against election deepfakes. Il Foglio's own call for a moratorium on deepfakes targeting opponents during the campaign has reportedly been endorsed by Giorgia Meloni and Elly Schlein. PD Chamber Vice-President Anna Ascani says Italy is entering the campaign "empty-handed". Her bill would have let AGCOM promptly remove deepfake videos of candidates during the campaign, and the Chamber's majority rejected it.
Since then, figures from Lega (Gian Marco Centinaio), Forza Italia (Alessandro Cattaneo) and Fratelli d'Italia (Filini) have said they are open to further action. Il Foglio also reports that the government is weighing "strong moral suasion" rather than binding rules.
The instinct to protect voters from fabricated footage of candidates is right. The open question is which instrument fits.
The strongest case for binding rules
Ascani's argument deserves a fair hearing. A gentlemen's agreement binds only those who choose to sign it. Anonymous accounts, foreign actors and fringe campaigners are not parties to it. A deepfake released days before a vote can circulate widely before any court acts, and a correction published afterwards rarely reaches the same audience. A regulator that can order removal quickly, as her bill would have allowed AGCOM to do, addresses the timing problem that criminal law handles badly.
Ascani herself says the government's approach has a weakness: under the Butti bill's offence, guilt must first be established through a criminal trial. Il Foglio describes that bill as creating a "deepfaking" offence and requiring deepfakes to be clearly labelled. We could not confirm its enactment status. Italy's Law 132 of 23 September 2025 on artificial intelligence has been in force since 10 October 2025. We could not retrieve its full text, so we make no claim here about its criminal provisions or how they relate to the Butti bill.
Why speed and discretion are the real risks
The trouble with fast takedown power is that it is cheap to invoke and costly to get wrong. A candidate or party can label legitimate satire, a clipped quote or an unflattering but authentic video as a "deepfake". An authority with a short removal window, and platforms with a strong incentive to comply, will often remove first and examine later. In an election, a wrongly removed video is itself a distortion of the campaign.
The evidence on automated moderation is not reassuring. The EFF's October 2026 piece, "Social Media When Elections Are Close: Doing the Impossible Better", notes that leaked 2020 documents showed Meta's algorithms incorrectly deleted nonviolent Arabic-language content 77 percent of the time. It warns that lawmakers "should avoid promoting and passing legislation that effectively or explicitly mandates automated moderation systems." Short removal deadlines tend to produce exactly that.
The EFF also says this "is not inherently an 'AI' problem". Deceptive political content existed long before generative tools. A rule keyed to the technology rather than the harm will be both over- and under-inclusive: it catches parody, and it misses a cheaply edited clip that fools voters just as well.
What a proportionate package looks like
The parties' stated openness can be turned into something workable without giving up speech protections.
- Labelling over removal. Filini's point that AI videos should be labelled is the least restrictive tool available. Disclosure leaves speech intact and gives viewers context. Labelling duties should attach to the party that publishes or pays to promote the material, not to every intermediary.
- Narrow definition. Cover realistic fabrications of a real candidate's words or actions that are presented as authentic and intended to deceive. Exclude satire, parody and obvious commentary.
- Independent review with a fast lane. If AGCOM is to have a role, its orders should be reasoned, limited to clear cases, and open to immediate appeal. AGCOM already treats AI-related disinformation and deepfakes as a policy subject. Its Rapporto Intelligenza Artificiale 2026, published on 17 June 2026, includes a section on them. Expertise exists, but expertise is not the same as a mandate to remove political speech on short notice.
- Sunset and transparency. Any campaign-period rule should expire after the election. The regulator should publish the number of removals, their grounds and appeal outcomes.
Moral suasion is not nothing, but it is not accountable
The government's reported preference for moral suasion has real advantages. It is fast, flexible and avoids writing a badly drafted statute in a hurry. A voluntary moratorium among the major parties would also set a norm that is hard to break without reputational cost.
Its weaknesses are accountability and fairness. Informal pressure on platforms, with no published criteria and no right of appeal, can drive over-removal without anyone being answerable. Nor does it constrain anyone outside the political class. If the government wants to lean on platforms, it should say publicly what it is asking for and in which cases.
The practical conclusion
Italy should not enter the campaign empty-handed, and it should not enter it with an unreviewable takedown power either. The most defensible path is a party-level moratorium backed by a narrow labelling duty and a quick judicial route for the clear cases. The Butti bill's labelling element points in that direction, if it is enacted and kept proportionate. Italy can address election deception without building a mechanism that a future majority could aim at its critics.