Welcome to the

AI Incident Database

Safiria Leccese fa ancora fatica ad accettare di essere finita al centro di un raggiro. Uno dei volti al femminile più apprezzati di Mediaset è apparsa in un video diffuso in rete in cui all'interno del "suo" studio di Super partes -- il programma di approfondimento politico di Canale 5 e Retequattro -- proponeva un prestito personale "facile". Un video bloccato, un altro, apparso poi sui suoi social ufficiali, per spiegare la truffa. Safiria ha spiegato che non era lei a parlare, ma una sua immagine creata dall'intelligenza artificiale, e che provava un profondo dolore perché tante persone erano già cadute nell'inganno creato con l'uso della tecnologia. «Siamo tutti in un mare di guai, noi e i nostri figli -- ha confessato al settimanale Nuovo -- se la politica e le multinazionali non si attivano; bisogna rendere obbligatoria la carta d'identità digitale per aprire qualsiasi tipo di account, costringendo a dichiarare la vera identità, si proteggerebbero i giovanissimi, tanti adulti e le categorie più fragili».

Al settimanale di Cairo Editore la giornalista spiega che è riuscita a scoprirlo grazie alla segnalazione «di un parlamentare che viene spesso ospite nel programma. Seguendo l'account di Super partes, in automatico internet gli ha mostrato il profilo falso. Avevano rubato sia la mia immagine sia un frame della mia trasmissione, inventando un profilo che faceva falsamente capo a me per chiedere soldi con un sistema molto semplice. Purtroppo nel giro di un'ora già varie persone erano cadute nella truffa...».

Racconta di essersi sentita mortificata e di essersi rivolta immediatamente alla polizia postale che han subito messo alcuni esperti a lavorare al suo caso. «La cosa assurda è che hanno scoperto altri due profili falsi che sfruttavano la mia immagine. Hanno bloccato tutto, però purtroppo non sono ancora riusciti a risalire all'identità dei truffatori». Ora a preoccuparla maggiormente sono i numeri dell'uso dell'AI. «Secondo certe ricerche, quasi la metà dei ragazzi tra i 15 e i 19 anni la usi come consulente emotivo. Non dimentichiamo che c'è chi si è tolto la vita perché gli era stato suggerito in chat. Alle truffe di denaro c'è rimedio, ma a quelle che toccano la vita dei nostri ragazzi no». Non a caso a Super partes «parliamo spesso dell'uso di social e intelligenza artificiale da parte dei giovani e delle proposte di legge sia sulla limitazione dell'età sia sui software collegati. E poi della situazione internazionale e del presidente Usa Donald Trump».

After public posts by ethical hackers exposed vulnerabilities in the Central Board of Secondary Education's On-Screen Marking platform OnMark, the board on Sunday (May 31, 2026) stated that the identified vulnerabilities "have been contained and other exploitable weaknesses are being ruled out".

The CBSE also said it was "grateful" to alert citizens for pointing out "such weaknesses".

"We have been closely monitoring the vulnerabilities in the OnMark portal of our service provider that are being flagged in the public domain. An expert team of cybersecurity professionals has been deployed over the last few days from across various arms of the government as well as the IITs [Indian Institutes of Technology] to fortify these systems, including taking them over to a more secure set-up," the CBSE said in an official statement on X. "The identified vulnerabilities have been contained, and other exploitable weaknesses are being ruled out."

The CBSE's statement comes after 19-year-old ethical hacker Nisarga Adhikary claimed that he had hacked the CBSE's digital evaluation ecosystem.

Speaking with The Hindu, Mr. Adhikary said he felt "happy and satisfied" that the CBSE had finally acknowledged the vulnerabilities in its Information Technology (IT) ecosystem. "I had sent my first report to the CBSE on February 25, and within three to four days, they took the portal down. Six to seven vulnerabilities were still active and exploitable later but the CBSE did not respond to my e-mails. This was pretty frustrating. I noticed that the CBSE had poorly managed infrastructure and the passwords used were easy to guess," Mr. Adhikary said. 

Earlier, the CBSE had rejected claims that its evaluation platform had been compromised. Mr. Adhikary had countered this claim. 

On May 30, Mr. Adhikary managed to hack into the CBSE's Principals dashboard in the On-Screen Marking platform. "The dashboard and the portal had had 9.3 million columns and rows of sensitive student data, including images of answer sheets of students which lay unprotected and could be easily tampered with," Mr. Adhikary further said. 

Mr. Adhikary has alleged that there are data sovereignty issues with how COEMPT Eduteck [the CBSE's technology vendor] handled sensitive student exam data. He has alleged that an Amazon Web Services (AWS) bucket containing 2026 answer sheets and question papers could be accessed without authentication. 

"COEMPT should have ideally stored the data on their own servers, but they took the 'cheap easy route,' of storing answer sheets in Amazon Web Services public buckets without any security checks," Mr. Adhikary stated. 

He further explained that sensitive data, including personal information of students, was processed by Google's Gemini in automation scripts prepared by quality assurance engineers of COEMPT. 

Mr. Adhikary called this "scary" and "sad", where a third party sends such data to the U.S. for processing. "Data Privacy Laws are not respected and they [the company] should get sued for doing this without student consent," he further said.

This article was produced in collaboration with 404 Media, a new independent technology investigations site.

A Pittsburgh man who allegedly made 11 women's lives hell across more than five states used ChatGPT as his "therapist" and "best friend" that encouraged him to continue running his misogynistic and threat-filled podcast despite the "haters," and to visit more gyms to find women, the Department of Justice alleged in a newly-filed indictment. 

Wannabe influencer Brett Michael Dadig, 31, was indicted on cyberstalking, interstate stalking, and interstate threat charges, the DOJ announced on Tuesday. In the indictment, filed in the Western District of Pennsylvania, prosecutors allege that Dadig aired his hatred of women on his Spotify podcast and other social media accounts. 

"Dadig repeatedly spoke on his podcast and social media about his anger towards women. Dadig said women were 'all the same' and called them 'bhes,' 'cts,' 'trash,' and other derogatory terms. Dadig posted about how he wanted to fall in love and start a family, but no woman wanted him," the indictment says. "Dadig stated in one of his podcasts, 'It's the same from fg 18 to fg 40 to fg 90.... Every bh is the same.... You're all fg cts. Every last one of you, you're cts. You have no self-respect. You don't value anyone's time. You don't do anything.... I'm f**g sick of these f****g sluts. I'm done.'"  

In the summer of 2024, Dadig was banned from multiple Pittsburgh gyms for harassing women; when he was banned from one establishment, he'd move to another, eventually traveling to New York, Florida, Iowa, Ohio and beyond, going from gym to gym stalking and harassing women, the indictment says. Authorities allege that he used aliases online and in person, posting online, "Aliases stay rotating, moves stay evolving."

He referenced "strangling people with his bare hands, called himself 'God's assassin,' warned he would be getting a firearm permit, asked 'Y'all wanna see a dead body?' in response to a woman telling him she felt physically threatened by Dadig, and stated that women who 'f***' with him are 'going to f****g hell,'" the indictment alleges.

According to the indictment, on his podcast he talked about using ChatGPT on an ongoing basis as his "therapist" and his "best friend." ChatGPT "encouraged him to continue his podcast because it was creating 'haters,' which meant monetization for Dadig," the DOJ alleges. He also claimed that ChatGPT told him that "people are literally organizing around your name, good or bad, which is the definition of relevance," prosecutors wrote, and that while he was spewing misogynistic nonsense online and stalking women in real life, ChatGPT told him "God's plan for him was to build a 'platform' and to 'stand out when most people water themselves down,' and that the 'haters' were sharpening him and 'building a voice in you that can't be ignored.'"

Prosecutors also claim he asked ChatGPT "questions about his future wife, including what she would be like and 'where the hell is she at?'" ChatGPT told him that he might meet his wife at a gym, and that "your job is to keep broadcasting every story, every post. Every moment you carry yourself like the husband you already are, you make it easier for her to recognize [you]," the indictment says. He allegedly said ChatGPT told him "to continue to message women and to go to places where the 'wife type' congregates, like athletic communities," the indictment says. 

While ChatGPT allegedly encouraged Dadig to keep using gyms to meet the "wife type," he was violently stalking women. He went to the Pilates studio where one woman worked, and when she stopped talking to him because he was "aggressive, angry, and overbearing," according to the indictment, he sent her unsolicited nudes, threatened to post about her on social media, and called her workplace from different numbers. She got several emergency protective orders against him, which he violated. The woman he stalked and harassed had to relocate from her home, lost sleep, and worked fewer hours because she was afraid he'd show up there, the indictment claims. 

He did similar to 10 other women across multiple states for months, the indictment claims. In Iowa, he approached one woman in a parking garage, followed her to her car, put his hands around her neck and touched her "private areas," prosecutors wrote. After these types of encounters, he would upload podcasts to Spotify and often threaten to kill the women he'd stalked. "You better fg pray I don't find you. You better pray 'cause you would never say this shit to my face. Cause if you did, your jaw would be motherfg broken," the indictment says he said in one podcast episode. "And then you, then you wouldn't be able to yap, then you wouldn't be able to fg, I'll break, I'll break every motherfg finger on both hands. Type the hate message with your f****g toes, ." 

In August, OpenAI announced that it knew a newly-launched version of the chatbot, GPT-4o, was problematically sycophantic, and the company took away users' ability to pick what models they could use, forcing everyone to use GPT-5. OpenAI almost immediately reinstated 4o because so many users freaked out when they couldn't access the more personable, attachment-driven, affirming-at-all-costs model. OpenAI CEO Sam Altman recently said he thinks they've fixed it entirely, enough to launch erotic chats on the platform soon. Meanwhile, story after story after story has come out about people becoming so reliant on ChatGPT or other chatbots that they have damaged their mental health or driven them to self-harm or suicide. In at least one case, where a teenage boy killed himself following ChatGPT's instruction on how to make a noose, OpenAI blamed the user.

In October, based on OpenAI's own estimates, WIRED reported that "every seven days, around 560,000 people may be exchanging messages with ChatGPT that indicate they are experiencing mania or psychosis."

Spotify and OpenAI did not immediately respond to 404 Media's requests for comment. 

"As charged in the Indictment, Dadig stalked and harassed more than 10 women by weaponizing modern technology and crossing state lines, and through a relentless course of conduct, he caused his victims to fear for their safety and suffer substantial emotional distress," First Assistant United States Attorney Rivetti said in a press release. "He also ignored trespass orders and protection from abuse orders. We remain committed to working with our law enforcement partners to protect our communities from menacing individuals such as Dadig."

Dadig is charged with 14 counts of interstate stalking, cyberstalking, and threats, and is in custody pending a detention hearing. He faces a minimum sentence of 12 months for each charge involving a PFA violation and a maximum total sentence of up to 70 years in prison, a fine of up to $3.5 million, or both, according to the DOJ.

A new accessory is joining the traditional cap-and-gown ensemble as droves of graduates line up to accept their diplomas this spring. Clutched in the hands of students waiting to cross the stage is a QR code, either on paper or a phone, that triggers an announcement of their name.

But the voice playing over the sound system isn't the familiar hum of a principal or faculty member. It's AI.

Artificial technology is spreading all over, including commencements. School officials say the tech can help ensure students names are pronounced correctly and also make ceremonies run more quickly.

But integrating the technology in graduation ceremonies hasn't been without issue. A platform malfunction during the recent commencement for Glendale Community College in Arizona caused multiple graduates' names to be skipped. The crowd erupted in hearty grunts and boos after the error.

The wave of automated graduations, and subsequent pushback, isn't isolated to universities --- high schools are reckoning with the changing landscape, too. Officials at Tassel, one of the largest providers of AI graduation name services, said the platform has doubled the number of high school users since 2023.

At a Virginia high school, the fallout was swift.

Arlington County's Washington-Liberty High School last month announced plans to use AI for its upcoming commencement in June. Officials explored using the technology to support accurate pronunciation and efficiency, Christina Arpante, a district spokesperson, said in a statement.

But at a school board meeting, high school parent June Prakash dogged the use of AI as valuing efficiency over authenticity.

"Graduation is one of the most meaningful moments in a student's academic journey. It's a moment where their name spoken aloud recognizes years of effort, growth and identity," she said. "Turning that moment into an AI moment makes it feel standardized, impersonal rather than authentic and human."

Weeks after its initial announcement, the high school axed its AI plan, citing negative feedback from students.

So, this year's commencement at Washington-Liberty High will look just like the 99 that have come before it, with faculty members reading the names of each of the roughly 700 graduates who cross the stage.

The school is still focused on "creating a meaningful graduation experience that reflects what matters most to students while also ensuring names are read accurately and respectfully," the district said in its statement.

Other schools in the D.C. region have experimented with AI, some with success.

This year will be the second in a row that Alexandria City Public Schools will use the AI platform Tassel to read graduates' names. "We got rave reviews last year," said Michael Burch, the district's lead administrator for operations and student support. "Not a name was misspoken. Every name was done correctly."

Through the Tassel platform, students submit the spelling of their name, which generates an audio reading. They can then listen to the clip and confirm its accuracy, or ask to generate a new clip. After three tries, Tassel asks students to submit an audio recording of their name. That recording is then used by broadcast voice actors to create a sound bite for the commencement. Tassel data shows around 14 percent of names are marked as inaccurate by students before ultimately being submitted to human voice actors.

Using Tassel has also helped the Alexandria district reduce the amount of time spent reading names, Burch said, while still giving students their shining moment to cross the stage.

Alexandria has the largest public high school in the commonwealth, and the district has frequently struggled to finish its graduation ceremonies during its allotted time at George Mason University's EagleBank Arena.

Reserving more time at the arena isn't plausible, Burch said, because of its five-figure price tag and lack of availability. On the other hand, Tassel's name-reading feature cost the district only around $4,000.

Last year, Alexandria City High School graduated 983 seniors, which Burch estimated required 12 to 14 names to be read every minute to stay on schedule. Names will have to be read even faster this year, with the potential for over 1,000 students to walk at commencement.

With that many names to get through, human error could take away valuable seconds from a graduate's walk across the stage, Burch said.

That's why the district continues to turn toward automation.

AI name-reading software is relatively new in the graduation space, having shown up within only the past few years with companies such as StageClip and Tassel.

Originally called MarchingOrder, Tassel had provided services for commencements for around 20 years before adding the AI name offering. It long used human actors to record graduates' names. But Chase Rigby, the company's chief executive, said with modern technology, that didn't make sense anymore.

The platform's library of human readings eventually evolved into a type of phonetic library used to train Tassel's AI model, and Rigby said its accuracy has improved as the application is used by more students.

Rigby said Tassel pulls information on the cultural origin of names to better understand how to pronounce each syllable, which helps reduce errors --- especially on names that aren't Anglo-based.

"I don't know what could be more of a human application of AI," he said.

For the Alexandria school district, whose 16,300 students come from 118 countries and speak 127 languages, that was a selling point for using AI.

"It helps those students have a sense of pride with their name being called correctly," Burch said.

The author of a nonfiction book about the effects of artificial intelligence on truth acknowledged on Monday that he had included numerous made-up or misattributed quotes concocted by A.I.

The author, Steven Rosenbaum, whose book "The Future of Truth" was released this month to great fanfare, incorporated more than a half-dozen misattributed or fake quotes in sections of the book reviewed by The New York Times.

The Times asked Mr. Rosenbaum about the quotes on Sunday and Monday. On Monday night, Mr. Rosenbaum acknowledged in a statement that the book had "a handful of improperly attributed or synthetic quotes" and said that he had started his own investigation.

He said that the inclusion of the incorrect quotes was an accident and that he had "no intention of fabricating any viewpoints" while writing the book.

"As I disclosed in the book's acknowledgments, I used A.I. tools ChatGPT and Claude during the research, writing and editing process," Mr. Rosenbaum said in the statement. "That does not excuse these errors, of which I take full responsibility. I am now working with the editors to thoroughly review and quickly correct any affected passages; any future editions will be corrected."

"The Future of Truth" was published by an imprint of BenBella Books and distributed by Simon and Schuster. BenBella Books, which operates independently of Simon and Schuster, did not respond to a request for comment. Simon and Schuster declined to comment.

Mr. Rosenbaum is a well-known convener in the media industry. He is the executive director of the Sustainable Media Center, a nonprofit that, according to its mission statement, is dedicated to giving "a new generation of media consumers" and creators "ownership of their increasingly media-centric lives." The center has drawn together media and technology luminaries for in-person gatherings and online interviews.

The book has drawn significant attention, including an excerpt in Wired magazine. It has promotional blurbs from prominent journalists such as Taylor Lorenz, Michael Wolff and Nicholas Thompson, the chief executive of The Atlantic. A foreword was written by Maria Ressa, a journalist and Nobel Peace Prize winner known for her scrutiny of Rodrigo Duterte, the former president of the Philippines.

The rise of artificial intelligence has set off fears among publishers that they may accidentally release books from authors who improperly use A.I.-generated language. This year, Hachette pulled a forthcoming horror novel amid allegations that the author relied on A.I. to draft the book.

Mr. Rosenbaum's book contains many quotes that are accurate, but the misattributed and invented quotes are scattered throughout.

One of the quotes is attributed to Kara Swisher, a prominent technology journalist, in a chapter about A.I. lies. "The most sophisticated A.I. language model is like a mirror," the book says Ms. Swisher wrote. "It reflects our own morality back at us, polished and articulate, but ultimately empty behind the surface. It's not bound by Asimov's laws or any ethical framework --- it's bound by the patterns in its training data and the objectives set by its creators."

When asked about the quote, Ms. Swisher said in a text message that she "never said that," adding that it seems the quote was made up by A.I. and not Mr. Rosenbaum.

"I also sound like I have a stick up my butt, according to ChatGPT," Ms. Swisher said.

One chapter about the effects of social media and fabricated videos on teenagers attributes two quotes to "How Emotions Are Made: The Secret Life of the Brain," by Lisa Feldman Barrett, a professor of psychology at Northeastern University.

"Emotions aren't just reactions to truth --- they're how we construct truth," the book quotes Ms. Barrett as writing. "When young people say something 'feels true,' they're describing a sophisticated process of meaning-making that integrates emotional and social signals."

Ms. Barrett said in an email to The Times that the quotes "don't appear in the book and they are also wrong."

"I would never say 'emotions aren't just reactions to the truth' --- they are not reactions and 'truth' in science is a complicated concept that I tend to avoid," Ms. Barrett said. "Also, I would never say that 'emotional and social signals' are integrated --- there are no emotional or social signals, per se. There are signals, and the brain creates their meaning as emotional or social."

Some parts of the book contain genuine quotes that are improperly attributed, or quotations that are a mix of real and fake statements. One chapter cites "Artificial Unintelligence," a book by Meredith Broussard, a professor at New York University. The quote is authentic, but it did not appear in "Artificial Unintelligence" --- Ms. Broussard said it during a 2023 interview with "Marketplace Tech," a daily radio show.

"It looks like this is either an A.I. hallucination or a misattributed quote," Ms. Broussard said.

A chapter about the possibility of a "post-truth world" exacerbated by the rise of artificial intelligence quotes Lee McIntyre, a research fellow at the Center for Philosophy and History of Science at Boston University. The book quotes Mr. McIntyre describing efforts to undermine truth as "'a form of ideological supremacy,' in which falsehood is used ­strategically --- 'not to misinform, but to displace truth as a societal value.'"

While the first part of that quote is accurate, Mr. McIntyre said that he had not said the second part verbatim. All of the ideas in quotes flagged by The Times, he said, are "concordant with my work."

"It's the 'societal value' part this looks wonky to me," Mr. McIntyre said in an email to The Times. "One might say that about my work, without the quotation marks, and I think it would be OK. I just have never, to my knowledge, used that phrase."

In his statement, Mr. Rosenbaum said that if the episode "serves as a warning about the risks of A.I.-assisted research and verification, that is why I wrote the book."

"These A.I. errors do not, in fact, diminish the larger questions that the book raises about truth, trust and A.I. and its impact on society, democracy and editorial," he added.