Italian far-right party publishes AI-generated racist images

What happened

Over the course of a month, dozens of AI-generated images appeared on Liga's social media accounts including Facebook, Instagram, and X. The images often depicted men of darker skin tone, frequently shown as armed, attacking women or police officers.

Liga's spokeswoman confirmed that some images on their social media accounts were 'digitally generated,' but stated that all posts were based on actual reports from Italian newspapers with names, dates and locations. In one example cited, Liga posted an image with the caption 'Stranger attacked conductor' using an AI-generated photo of a man of darker skin tone with raised fist, while the original Il Resto del Carlino article only described the perpetrator as 'a stranger' without providing a photograph.

In another case, Liga posted an AI-generated image showing parents in Islamic clothing shouting at a girl, citing an Il Giorno report about child abuse. The original report made no mention of the family's religion and published no photograph of the family.

Reported impact

Opposition parties including the Partito Democratico and the Alliance of Greens and Left characterized the images as containing racist, Islamophobic, and xenophobic hate speech. Democratic Party Senator Antonio Nicita stated the images 'contain almost all forms of hate speech—from racism and xenophobia to Islamophobia' and characterized them as 'violent and deceptive,' as the blurring of victims' faces creates the impression the photographs are real.

Francesco Emilio Borrelli of the Greens and Left Alliance described the practice as part of a strategy to spread fear among citizens, stating 'AI generates content according to given instructions, and here it is clear that orders were given to create images of Black people robbing elderly women or frightening citizens.'

Index analysis

Researcher Salvatore Romano from the nonprofit AI Forensics identified telltale signs of AI generation in the images, noting that 'the subject is in the foreground while the background is blurred, which is a typical sign of AI-generated content,' and expressed concern that such images are becoming increasingly convincing.

The complaint to Agcom noted examples of images published with logos of reputable media outlets, even though those outlets had not published photographs of the alleged perpetrators.

Classification

Organizations

Liga (subject), Matteo Salvini (subject), Partito Democratico, Agcom (regulator)

Category

Harmful and abusive content

Industry

Media, social platforms and marketing

Country

Italy

Harm status

Potential harm

Intent

Intentional

Timing

Unknown

Harm types

Human or fundamental rights, Reputational, Psychological

Harmed

General public, Civil society

Failure modes

Deepfake image, Hallucination

Consequences

Regulatory investigation

AI systems

AI image generation tool

Sources and evidence