Skip to main content
NOPE
Low reported severity Internal Action

Genoa corruption inquiry: press reports say speech-to-text software turned a suspect's 'lawful financing' into 'unlawful' in his interrogation transcript

In May 2024 a businessman questioned as a suspect in the Genoa prosecutors' inquiry into alleged bribes to the president of the Liguria region was recorded in the official transcript of his interrogation before the preliminary-investigations judge as saying that the president wanted 'finanziamenti illeciti' (unlawful financing). According to Corriere della Sera, on receiving the transcript he wrote to the judge to contest it, saying he had said 'finanziamenti leciti' (lawful financing); his lawyer said the legally valid handwritten record did not contain the word. Il Post reports that the passage had been highlighted in the headlines of many newspapers in the following days as confirming the prosecutors' accusations, and that a second hearing of the audio was needed to settle it. La Repubblica reported on 27 May 2024 that, at an ad hoc hearing, the businessman confirmed he had said 'leciti', with the prosecutor's agreement according to his lawyer, and the judge took note; it wrote that the fault appears to lie with the automatic software that transcribes words spoken in court. Il Post reports that checks found the error was not human and that the transcript had been produced by speech-to-text software. The software is not named.

AI system
Unidentified speech-to-text tool
Occurred
May 2024
Reported
18 May 2024
Event location
Italy
What the AI did
Made a claim about the person
Reported harm
Reputational Harm
Whose AI use
An institution’s AI use
Setting
Justice
Evidence
AI involvement reported · Causal attribution supported · 3 sources
6 claims: 6 reported. 3 open questions
People reported harmed
1 person

AI system as recorded: Unnamed speech-to-text software that, according to Il Post, produced the transcript of the interrogation attached to the inquiry file; la Repubblica describes it as automatic software that writes down words spoken in court

Reported harm

According to Il Post, an automatic transcript wrongly recorded a suspect as saying the regional president wanted unlawful rather than lawful financing, and many newspapers headlined the passage as confirming the accusations until, at an ad hoc hearing reported by la Repubblica, he confirmed he had said 'lawful' and the judge took note.

Outcome

Resolved

At an ad hoc hearing in late May 2024 the businessman confirmed he had said 'leciti' and the preliminary-investigations judge took note (la Repubblica, 27 May 2024, relaying the defence lawyer).

What remains unknown

  • Which speech-to-text product the court registry used and who carried out the checks Il Post mentions.
  • The exact date of the correction hearing: Il Post gives 26 May 2024, a Sunday, while la Repubblica of 27 May 2024 says it took place that morning.
  • Whether the error had any effect on the proceedings beyond the press coverage and the extra hearing.

What the evidence supports

AI involvement: reported. Il Post reports that checks found the error in the interrogation transcript was not human and that the transcript had been produced by speech-to-text software, which rendered the suspect's 'leciti' as 'illeciti'; la Repubblica writes that the fault appears to lie with the automatic transcription software. That transcribed word is the statement newspapers headlined, according to Il Post, until an ad hoc hearing established that he had said 'leciti'. The software is not named, its workings are described only in general terms, and la Repubblica notes the pronunciation was apparently doubtful.

6 claims: 6 reported. What the statuses mean

Reported The official transcript of the suspect's interrogation before the Genoa preliminary-investigations judge recorded him as saying the regional president wanted 'finanziamenti illeciti' (unlawful financing).

Causal attribution. Contents of the transcript as the outlets report them; the transcript itself was not inspected.

Reported On receiving the transcript the suspect wrote to the judge contesting it and saying he had said 'leciti' (lawful); his lawyer said the handwritten record drawn up by the registrar did not contain the word 'illeciti'.

Causal attribution. The suspect's and his lawyer's account.

  • corriere.it(opens in new tab) supports · Italian
    'contesta recisamente» la trascrizione ufficiale perché ricorda «con assoluta certezza» di aver detto «finanziamenti leciti»'; 'Infatti, la frase incriminata non c’è nel verbale stilato dal cancelliere.'
Reported According to Il Post, the passage had been highlighted in the headlines of many newspapers in the following days because the single adjective would have unequivocally confirmed the prosecutors' accusations.

Causal attribution. Il Post's account of the press coverage that followed the transcript; it links the coverage to the transcribed word.

  • ilpost.it(opens in new tab) supports · Italian
    'Un solo aggettivo – illeciti – avrebbe confermato in modo inequivocabile le accuse della procura, e non a caso proprio quel passaggio era stato messo in evidenza nei titoli di molti giornali nei giorni successivi'
Reported At an ad hoc hearing the suspect confirmed he had said 'leciti', which according to his lawyer the prosecutor accepted, and the judge took note.

Causal attribution. The defence lawyer's statement relayed by la Repubblica; Il Post reports that a second hearing of the audio settled the doubt.

  • genova.repubblica.it(opens in new tab) supports · Italian
    'lo stesso imprenditore ha chiarito “di aver pronunciato la parola leciti, lo abbiamo anche concordato con il pubblico ministero”'; 'A questo punto la gip ha preso atto.'
  • ilpost.it(opens in new tab) supports · Italian
    'si è dovuto risolvere il dubbio con un secondo ascolto'
Reported Il Post reports that checks found the error was not human and that the transcript attached to the inquiry had been produced by speech-to-text software; la Repubblica writes that the fault appears to lie with the automatic software that transcribes words spoken in court.

Causal attribution. Attribution of the error to the speech-to-text software rests on Il Post's account of the checks (whose author is not named) and la Repubblica's hedged inference; the software is not named.

  • ilpost.it(opens in new tab) supports · Italian
    'l’origine di questa svista non era dovuta a un errore umano'; 'era stata fatta da un software speech-to-text'
  • genova.repubblica.it(opens in new tab) supports · Italian
    'la “colpa” pare essere del software automatico che riporta nero su bianco le parole pronunciate in udienza'
Reported La Repubblica describes the suspect's pronunciation as apparently doubtful, and Il Post reports that the registrar summarising the interrogation had not mentioned the passage at all.

Causal attribution. Context and counterevidence: unclear speech and the human summary's silence bear on how the error arose.

Sources

3 sources inspected. Sources that repeat one account do not corroborate each other.

How the sources were read, and where the events happened

Read in Italian on 2026-10-25 (full article). Translation by the research agent, an AI. Applies to s1, s3.

Read in Italian on 2026-10-25 (full article, also present as JSON-LD articleBody). Translation by the research agent, an AI. Applies to s2.

Event countries: Italy. Affected-person countries: Unknown. Court countries: Italy.

The interrogation and the hearing that corrected the transcript took place before the preliminary-investigations judge in Genoa, as Il Post and la Repubblica describe; no source states where the suspect lives.

Reviewed for publication 2026-10-25: Published as an automated recognition error about a person in a criminal inquiry: Il Post and la Repubblica attribute to speech-to-text software a transcript that recorded a suspect saying the opposite of his words, which, Il Post reports, newspapers headlined as confirming the accusations until an ad hoc hearing corrected it.

People described

A businessman questioned as a suspect in a 2024 corruption inquiry in Genoa

People reported harmed in this case

1 person

0 AI participants · 1 other person harmed

One person: the suspect whose words the transcript misrecorded. Newspaper readers are an audience and are not counted.

Counted once within this case. The same person may appear in other cases. This count does not establish AI causation.

Tags

#speech-to-text#court-transcript#criminal-investigation#misattributed-statement

Cite this case

Compiled per our published methodology: verification statuses, sourcing standards, and corrections process.

APA

NOPE. (2026). Genoa corruption inquiry: press reports say speech-to-text software turned a suspect's 'lawful financing' into 'unlawful' in his interrogation transcript. AI incidents. https://nope.net/incidents/2024-genoa-corruption-inquiry-speech-to-text-transcript-recorded-suspects-lawful-financing-as-unlawful

BibTeX

@misc{2024_genoa_corruption_inquiry_speech_to_text_transcript_recorded_suspects_lawful_financing_as_unlawful,
  title = {Genoa corruption inquiry: press reports say speech-to-text software turned a suspect's 'lawful financing' into 'unlawful' in his interrogation transcript},
  author = {NOPE},
  year = {2026},
  howpublished = {AI incidents},
  url = {https://nope.net/incidents/2024-genoa-corruption-inquiry-speech-to-text-transcript-recorded-suspects-lawful-financing-as-unlawful}
}

Related cases

Medium Google AI Mode

German court bars Google from repeating an AI Mode answer that falsely called a doctor a convicted sex offender

In late August 2026 a colleague told a German physician, until that summer a chief physician at a hospital, that searching his name with the word for suspension in Google's AI Mode produced what read like a factual report: that he had been suspended and dismissed and had received a suspended sentence for sexual coercion. According to legal news site beck-aktuell, none of it concerned him; the AI had apparently attached facts about another former senior employee of the same hospital to his name, and the newspaper articles it linked as sources did not name him. Google said it had reviewed his complaint and would take no action, and a lawyer's warning letter also failed. On 24 September 2026 the Aachen Regional Court issued a preliminary injunction (8 O 268/26) barring Google Ireland Limited from spreading the claims through AI Mode, finding that they violated his personality rights and that Google is responsible for its AI search answers.

High Grok

Grok falsely named former officer Christi Hill as an arresting officer in the Henry Nowak case; she is in hiding after threats

In early June 2026, social media posts and the AI chatbot Grok falsely identified former Hampshire police constable Christi Hill as one of the 'primary officers shown' in bodycam footage of the December 2025 arrest of Henry Nowak, who died after being stabbed. Hill had left the force in April 2024 — more than a year before the murder — and says the misidentification stems from a national police bravery award photo being misattributed. After 'endless threats' and 'threats of violence' she is hiding in a safe house. BBC Verify separately obtained evidence that serving officer Tristan Parsons, also wrongly accused online, was not in the country at the time; the Home Secretary said a misidentified male officer had to move out of his home. Hampshire Constabulary confirmed an unrelated officer was misidentified and received death threats.

High Clearview AI facial recognition (reported)

Tennessee grandmother Angela Lipps jailed five months on an AI facial-recognition misidentification; charges dismissed after bank records placed her 1,200 miles away

On 14 July 2025, U.S. Marshals arrested Angela Lipps, a 50-year-old Tennessee grandmother, at gunpoint at her home while she was babysitting four children, on a nationwide-extradition North Dakota warrant: West Fargo police's AI facial-recognition technology had tied her to bank-fraud surveillance of a woman using a fake U.S. Army military ID to withdraw tens of thousands of dollars. She fought extradition for about three months in a Tennessee jail, was booked into Cass County, North Dakota on 30 October 2025 on four counts of unauthorized use of personal identifying information and four counts of theft, and was released on 24 December after her bank records showed her more than 1,200 miles away in Tennessee; charges were dismissed without prejudice. Fargo's police chief acknowledged 'incorrect assumptions' linked to the AI identification; an email obtained by KVLY shows six Fargo detectives were notified of her arrest in July, nearly five months before the department says it knew she was in custody. She lost her home, her car and her dog; her attorneys are pursuing potential civil-rights claims, with no lawsuit filed as of 31 March 2026. On 15 September 2026 she filed a $10 million federal civil-rights suit in the District of North Dakota against the City of Fargo and the detective, alleging he relied on a flawed facial-recognition match and ignored exculpatory evidence.

Medium Google AI Overviews

Keene v. Google: defamation claim over AI Overviews calling an author a convicted murderer survives dismissal

An author and television producer, the subject of the Apple TV series 'Black Bird', sued Google, alleging that AI Overviews shown for searches about him in May and June 2025 falsely said he was serving a life sentence without parole, in one version for the murders of three women. He says he notified Google three times and received apologies attributing the statements to its AI, yet similar statements kept appearing. On 14 September 2026 the Northern District of Illinois denied Google's motion to dismiss as to two of the four challenged Overviews and dismissed the other two without prejudice. This was a ruling on the pleadings, not a finding that the allegations were proven.

If you or someone you know is struggling, free and confidential support is available. Find a helpline near you at Signpost.

Back to all incidents