Italy's PM was deepfaked in lingerie. She could defend herself. Most people can't.

The Pattern

In May 2026, Italian Prime Minister Giorgia Meloni posted on X that several fake images of her, created with AI and passed off as real, were circulating on social media. One showed her seated on a bed in lace underwear. The fabricated image went viral. Some users who appeared to believe it was genuine commented that such an image was “shameful” and unbecoming of a prime minister.

Meloni’s response was notable for what it acknowledged about power. “I must admit that whoever created them, at least in the attached case, has also improved me quite a bit,” she wrote. “But the fact remains that nowadays anything is used to attack and make up falsehoods about me. The point, however, goes beyond me. Deepfakes are a dangerous tool, because they can deceive, manipulate, and hit anyone. I can defend myself. Many others cannot.”

The asymmetry of defense

Meloni has a platform, a staff, and a global audience. When she says an image is fake, people listen. When a private citizen is deepfaked, the context does not matter. Lingerie, compromising positions, situations they were never in. They do not have a press office. They do not have millions of followers. They do not have the ability to make “this is fake” a headline. They have a report button and a hope that the platform responds before the image reaches their employer, their family, or their community.

This is the structural problem with AI-generated harassment: the cost of creation is approaching zero, while the cost of defense remains high. Anyone with a smartphone can generate a convincing deepfake. The target needs forensic analysis, legal counsel, and platform cooperation to disprove it. The asymmetry favors the attacker in every case where the target is not a head of government.

The Grok image problem

The Meloni deepfakes emerged in the same period that X’s Grok chatbot was criticized for its image editing modes, which were used to generate non-consensual sexualized imagery of women and children. Grok’s image generation was eventually restricted after widespread criticism, but the underlying tools remain available across dozens of platforms. AFP reported in France24 that advanced AI visual generators have “largely erased the once-telltale glitch of extra fingers” and can now churn out uncannily real-looking deepfakes within seconds. The defensive tools have not kept pace. AI detection, watermarking, platform moderation. They are all behind.

The liar’s dividend for individuals

The Meloni case also demonstrates the liar’s dividend at the individual level. Once a deepfake of a public figure goes viral, every real image of that person becomes suspect. The audience that saw the fake does not necessarily see the correction. The audience that knows deepfakes exist begins to question real images. The result is a general erosion of trust in visual evidence, not just for the person who was deepfaked, but for everyone.

Meloni’s final message was: “Check before believing, and believe before sharing. Because today it has happened to me, tomorrow it can happen to anyone.” That is the correct advice. It is also advice that puts the entire burden on the audience, none on the platforms that host and amplify the fakes, and none on the companies that build the tools that generate them. The prime minister can say that and be heard. The private citizen who is deepfaked tomorrow cannot.

Verdict: Fake. The images were AI-generated. Meloni confirmed it. The deeper verdict is about the system: the tools that made the fakes are freely available, the platforms that amplified them face no consequences, and the people who cannot defend themselves have no recourse. That is the propaganda: not the image itself, but the infrastructure that makes the image possible and the asymmetry that makes it effective.