After the indignation caused in an American school by the broadcast of racist remarks attributed to the principal, the dizziness of discovering that the soundtrack was put together from scratch. This episode illustrates the dangers of artificial intelligence becoming accessible to everyone.
• Read also: Hyper-fake racist remarks: he uses AI to take revenge
Eric Eiswert, principal in Pikesville, Maryland, near Washington, found himself at the heart of a violent controversy with a voice recording – which turned out to be fake – of him making shocking comments against Jewish students and “ungrateful black children”.
The case, which occurs in the middle of an election year in the United States, highlights the ease with which generative AI tools can be used to harm anyone, and the difficulties authorities face in combating such practices.
“Now everyone is vulnerable,” and not just celebrities, warns Hany Farid, professor at the University of California at Berkeley (west).
“It only takes one image to add a person to a video, and 30 seconds of audio to clone someone’s voice,” continues the specialist in detecting digitally manipulated images and recordings, consulted by the police. in that case.
When the recording leaked onto social media in January, it quickly went viral. A publication collects thousands of comments on Instagram, and propels the school into the heart of a national controversy.
Civil rights activist DeRay McKesson calls for the principal’s resignation on his X account, followed by nearly a million Internet users. He will admit to being abused.
Hateful messages are raining down on social networks and threatening phone calls are multiplying in the establishment. The “world would be better if you were underground,” writes one Internet user to the principal.
The latter is placed on leave, his home placed under protection. Contacted by AFP, he did not respond.
Revenge
“I continue to worry about the damage caused by this affair,” confides Billy Burke, director of the union representing the principal.
At the end of April, Dazhon Darien, 31, the school’s sports manager, was arrested by authorities, accused of being behind the forgery. Investigators traced him to the email address that initially shared the file.
He would have acted to take revenge for an investigation opened against him by the principal into suspicious payments.
The defendant conducted research on AI tools from the school’s computer network, according to the indictment.
The soundtrack, according to the analysis of an expert consulted by the police, “contains traces of content generated by AI, with a posteriori human editing”.
This case demonstrates the need to “adapt the law to technological advances,” said local prosecutor Scott Shellenberger.
Many American schools are already having to fight against the distribution by students of pornographic images of their classmates made using generative AI, a practice that has caused terrible waves of harassment.
Audio montages are particularly difficult to detect. In January, a message broadcast by automated telephone calls spoofing the voice of President Joe Biden encouraged Democratic voters in the state of New Hampshire (northeast) to abstain in the party’s primaries.
In Pikeville, the affair shook the residents, “very close to each other,” says Parker Bratton, the high school golf coach.
“There is only one president, but there are a million principals!”, he worries, “people ask themselves: ‘What will happen to me if someone simply decides to destroy my career?”
