AI-made speeches and messages that sparked controversy

"Now improve the presentation and reasoning of this paragraph." The sentence, typical of someone talking to an AI tool, appeared in February 2026 in the opinion of an appellate judge at the Minas Gerais Court of Justice, according to Poder360. Internet users spread the excerpt, lawmakers filed a complaint with the National Council of Justice and the case joined other texts by authorities in which AI was exposed.

In this third article of the series, we gather episodes documented by news outlets in Brazil and abroad. Several of them involve laws, court opinions and posts, genres that differ from the podium speech. In all of them, an authority presented a text or a speech to the public, and the way that material was made ended up exposed. We attribute each fact to the report that recorded it, with the caveats it includes.

The Cuiabá councilor and the ChatGPT sentence in his post

In October 2025, city councilor Dídimo Vovô (PSB), of Cuiabá, published a text criticizing Mayor Abílio Brunini's (PL) trip to Dubai and China. The post began with the sentence ChatGPT places at the top of its answers: "Here is the revised and improved version of your text." According to FolhaMax, the slip became a topic on social media. One comment recorded in the report said the councilor "forgot to edit the caption ChatGPT created." Another pointed out that deleting it would not help: "People already took screenshots!"

The councilor deleted the post and published it again without the sentence. An aide admitted the mistake in an Instagram video embedded in the article and defended the tool, saying: "It gives you mobility, even though I used it the wrong way." FolhaMax is a regional outlet and does not report reach figures. Even with that limitation, it is the documented Brazilian case closest to an AI text that became a public joke.

The forgotten AI command in a TJMG appellate judge's opinion

The Minas Gerais case carries more weight. On February 11, 2026, the 9th Criminal Chamber of the TJMG acquitted, by majority vote, a 35-year-old man accused of the statutory rape of a 12-year-old girl. The opinion of the rapporteur, appellate judge Magid Nauef Láuar, according to Poder360, contained the command addressed to AI in the middle of the text.

According to Poder360 and the newspaper Correio, federal deputy Nikolas Ferreira (PL-MG) posted a video criticizing the lack of care, lawmakers filed a complaint with the CNJ and the Minas Gerais Public Prosecutor's Office said it would review the decision. The two reports differ in details: Correio transcribes the sentence with a slightly different wording and does not name the judge, whose name appears in Poder360, which uses the other wording. Neither reports figures on the reach of the reaction.

Laws and rulings made with AI: Porto Alegre, Acre and Peru

In 2023, the Porto Alegre City Council unanimously passed, and Mayor Sebastião Melo (MDB) signed, a law that exempts property owners from paying for a new water meter when the old one is stolen. According to Canaltech, the text came out of ChatGPT, with eight articles and the justification, and the tool even added an article no one had asked for. The author, councilor Ramiro Rosário (PSDB), only revealed the use after the law was signed. The council president, Hamilton Sossmeier (PTB), who voted without knowing where the text came from, said the case "turns on a warning light" and called the episode "a dangerous, complicated precedent."

In November 2023, Olhar Digital, based on a report by g1, reported that federal judge Jefferson Ferreira Rodrigues, of Acre, signed a ruling that cited Superior Court of Justice decisions that do not exist, found with ChatGPT. The CNJ opened an investigation. The judge treated the case as a "routine mistake" and attributed the research to a member of his office staff.

In Peru, in May 2024, congressman Paul Gutiérrez Ticona introduced a bill that included the AI's pasted answer, according to Exitosa, based on the newspaper El Comercio. In free translation, the text said: "As an AI language model, I do not have access to real-time information." The report does not record any response from the congressman or from others.

Heads of government and AI: Sweden and New York

In August 2025, Sweden's prime minister, Ulf Kristersson, admitted to using ChatGPT and LeChat to get a second opinion. "I use it myself quite often," he said, in a quote translated by Fast Company Brasil. An editorial in the newspaper Aftonbladet said he had "fallen into the oligarchs' AI psychosis," and researcher Virginia Dignum, of Umeå University, summed up the criticism: "We didn't vote for ChatGPT." The government spokesperson said the use served as a general reference and did not involve security-sensitive information. The episode concerns the use of AI in office and is unrelated to a specific speech.

In New York, in October 2023, automated calls from City Hall used the AI-generated voice of Mayor Eric Adams in Spanish, Yiddish, Mandarin, Cantonese and Haitian Creole. Adams speaks only English, and the calls did not disclose that the voice was synthetic. According to StateScoop, based on the AP, there were more than 5,000 calls in Spanish, more than 250 in Yiddish, more than 160 in Mandarin, 89 in Cantonese and 23 in Haitian Creole. Albert Fox Cahn, of the organization S.T.O.P., said the initiative was "deeply unethical, especially with taxpayer money" and called it a vanity project.

The pattern that links these cases to the podium speech

When we read the episodes together, it becomes clear under what circumstances AI becomes a public problem. In Cuiabá, Minas Gerais and Peru, it was discovered through carelessness, with a machine sentence left in the text. In Porto Alegre, it was revealed only after colleagues had voted. In New York, the technology made the mayor seem to speak languages he does not speak. In the cases where the reports record a reaction, the criticism turned toward the author and the trust he deserves.

This pattern connects with two other articles in the series. The fourth analyzes the expression "I rise to speak," which multiplied in speeches in the British Parliament and became a sign of AI-written text. The fifth covers the message sent by Vanderbilt after a shooting, in which the text's origin in ChatGPT led to a public apology.

Now think about your next event. An inauguration speech, a tribute or a convention opening is read live, often filmed and posted by the company itself. A sentence that sounds like a chatbot answer is recorded on video, and that video can circulate among employees, customers and competitors. It takes only a few seconds for someone to take a screenshot or clip the excerpt.