On May 6, 2026, Italian Prime Minister Giorgia Meloni did something unusual: she posted an AI-fabricated image of herself. The picture, which showed her in underwear and had circulated as if real, became her exhibit. "Check before you believe, and believe before you share," she wrote, warning that deepfakes "can deceive, manipulate and target anyone" (Euronews).
The irony is that weeks earlier, Meloni's own right-wing majority had rejected a bill aimed squarely at this problem. The Democratic Party proposal, led by deputy speaker Anna Ascani, would have banned the creation and dissemination of AI-manipulated political content during electoral and referendum campaigns, and — crucially — empowered communications regulator AGCOM to order platforms to remove that content within 24 hours and levy administrative fines for non-compliance (Il Sole 24 Ore).
The gap is real
Italy is not starting from zero. Law No. 132/2025, in force since October 10, 2025, added Article 612-quater to the Criminal Code, punishing anyone who disseminates AI-falsified images, video or audio that mislead about their authenticity and cause unjust harm — with prison terms of one to five years (Norton Rose Fulbright). Meloni herself relied on the broader legal toolkit in 2024, suing two men for €100,000 over fabricated pornographic videos.
But a criminal statute is a slow instrument. A prosecution unfolds over months; an election deepfake does its damage in hours. As Ascani put it, what Italy lacks is "a law that allows the authorities to request immediate removal" of content that "pollutes public debate and consensus formation, particularly in electoral campaigns." Criminal liability deters and punishes the creator after the fact. It does nothing to pull a viral fake off a feed before a vote.
The strongest case for the bill is timing: a fabricated clip of a candidate "confessing" the night before polls open can move results, and a 24-hour takedown order is the only tool fast enough to matter.
That is a serious argument, and it deserves to be met on the merits rather than waved away. Italy holds national elections by 2027, and the threat of synthetic media — domestic or foreign-sourced — is not hypothetical. A regulator that can compel removal within a day, backed by fines, plugs a genuine hole between the criminal code and the speed of the platform.
Why the rejection was nonetheless defensible
The majority's stated objection was not that the problem is fake. It was that the bill overlapped with norms already in a government delegation on AI and sat "partly in contrast with European regulation" (Andrea Casu). On the second point, the majority has a real argument. The EU Digital Services Act already obliges very large platforms to provide notice-and-action mechanisms and to mitigate systemic risks to electoral integrity, and the EU AI Act imposes transparency-labeling duties on deepfakes from August 2026. A national election-removal regime layered on top risks duplicative, and potentially conflicting, obligations — exactly the fragmentation the single market is meant to avoid.
The deeper concern is the bill's front end. Banning the creation of manipulated political content is constitutionally fraught: satire, parody, and clearly-labeled synthetic campaign ads are protected expression, and a creation ban invites platforms to over-remove anything that might draw a regulator's attention. The chilling effect of a fast, fine-backed takedown order points the same way. Faced with a 24-hour deadline and the threat of sanctions, a rational platform errs toward deletion — and legitimate political speech, the hardest category to adjudicate under time pressure, is the first casualty. Spain's and France's experiences with rapid election-content regimes show how easily "remove the fake" slides into "remove the contested."
The proportionate path
None of this argues for inaction. It argues for calibration. A defensible Italian regime would do four things the rejected bill conflated. First, drop the creation ban and target dissemination at scale of unlabeled, deceptive synthetic media — the actual harm. Second, build the takedown power as a narrow, election-window mechanism with a hard sunset and judicial review on appeal, not a standing censorship rail. Third, anchor it explicitly in the DSA's existing notice-and-action architecture so platforms face one coherent obligation, not two. Fourth, lead with transparency: mandatory provenance labeling — the AI Act's own approach — lets voters discount a fake without anyone deciding what they may see.
Meloni's instinct, "verify before believing," is the right civic reflex, and it is also the right regulatory philosophy. The state's job in an information ecosystem is to make manipulation legible and to give platforms clear, narrow duties — not to appoint a regulator as the arbiter of campaign truth on a 24-hour clock. Italy's gap between a slow criminal law and fast-moving fakes is real. The fix is a scalpel calibrated to the DSA, not the blunt instrument the Chamber voted down — or the silence that has followed it.