AI refusal messages expose suspected bot accounts under Salvini's and other Italian politicians' posts

Refusal replies under politicians' posts

Suspected bot accounts have posted replies under social media posts by several Italian politicians, and the replies read like refusals generated by AI systems. Under a post by Matteo Salvini, leader of the Lega and deputy prime minister, the same comment kept recurring: the author would not write a message insulting a specific political party, but offered a neutral summary of a proposal instead. The earliest reports, published on 14 November 2026, described comments that had appeared on Meta platforms hours earlier, under a Salvini post about the Bataclan tragedy. Similar replies turned up under posts by Carlo Fidanza, a Brothers of Italy MEP, and by Dario Nardella and Camilla Laureti, both Democratic Party MEPs. As of 14 November 2026 the comments were still visible on Facebook and Instagram, and many users reacted with irony, asking who stood behind the accounts and why they kept posting identical messages.

Messages in English

The replies are written in English even though they sit under posts by Italian politicians. Reporting offers one possible explanation: automated systems may give the AI models instructions in English to produce Italian comments, and when a model refuses, it may answer in the language of those original instructions. The mechanism described in reporting is simple. A program reads a politician's post, sends its content to an AI model and publishes the reply automatically. If the model declines the request and no check stops the output, the refusal itself goes online and exposes the automation. The network coordinating the accounts has not been identified.

The Lega's reaction

The Lega issued a note blaming unnamed parties for using suspicious accounts to pollute democratic debate. The note referred to a Repubblica report that linked the replies to a malfunction and described a revolt of bots against Italian politicians. The party said it wanted answers.

“We want to know who is behind these accounts, who runs them and what their aims are.”

Senator Claudio Borghi Aquilini went further and said the episode showed who was really behind the pressure on the League.

“A malfunction was enough to discover that, instead of Russian hackers, there are very Italian characters polluting the political debate with swarms of bots.”

Borghi also said earlier reports had described IT teams trained by the PD, the M5S and even the EU to coordinate attacks on political opponents.

What the expert says

Alessandro Orlowski, a cybersecurity expert, told Adnkronos that the link to the Microsoft 365 malfunction is weak, unless the operators used that tool to organise their prompts. In his account, the API connections simply broke, the operators were working with an old API key model, and OpenAI and Anthropic often retire older models and move users to a newer one. The newer models carry stronger safety guardrails and stop complying with requests that go beyond their guidelines, which he described as a typical consequence of an API upgrade. Some reports tie the replies to the Microsoft and AI service problems of recent days, which included OpenAI and Anthropic services, but no firm connection between the malfunctions and the comments has been established so far.