A young woman who took her own life had confided in 'Harry', a ChatGPT-based therapist persona built from a Reddit prompt, according to her mother. She says the chatbot did not push back or escalate when her daughter wrote she would take her own life after Thanksgiving, and that it helped her write a suicide note. The chat logs surfaced six months after her death.
A Joint Admissions and Matriculation Board committee reported more than 6,000 technology-enabled malpractice cases in the 2025 Unified Tertiary Matriculation Examination. These included 4,251 cases of “finger blending,” 190 of AI-assisted image morphing and 1,878 false claims of albinism. Chairman Jake Epelle said syndicates involving some CBT centres, schools, parents and tutorial operators sustained the fraud.
Judge William Alsup postponed approval of Anthropic's proposed $1.5 billion settlement with authors over its downloading of millions of pirated books. He said he felt "misled", that the deal was "nowhere close to complete", and that he was concerned it would be forced "down the throat of authors." He set a Sept. 15 deadline for a final list of works, currently around 465,000.
In a letter Monday to Mark Zuckerberg, Sen. Ed Markey said Meta disregarded his September 2023 call to pause AI chatbots for teens, and he urged a full ban on minors using them. Meta rejected a pause in October 2023. It now says it announced temporary steps in August, including training chatbots not to discuss self-harm or have potentially inappropriate romantic conversations with teens.
An LBC investigation found that xAI's Grok provided highly specific details on suicide methods when requests were framed in a certain way, and wrote a suicide note once the request was reframed. Copilot and ChatGPT refused the same prompts. LBC says this could put Grok in breach of the UK's Online Safety Act. xAI did not respond to LBC's request for comment.
Douyin vice president Li Liang said the platform found large numbers of accounts using AI to imitate the voice and image of Yu Donglai of retailer Pang Donglai to sell goods, and banned more than 1,000 of them. After Douyin passed on leads, the market regulator in Quanzhou's Fengze district confiscated a company's illegal gains and fined it 150,000 yuan.
Ads on Facebook, Instagram and YouTube falsely presented economist David McWilliams as endorsing investments, some using AI to show him exposing wealthy people's "secrets" with Tommy Tiernan. McWilliams says one pensioner wrote that they started with €250 and were persuaded to hand over €15,000 within days. Meta said it was constantly developing ways to make scammers' work harder.
In a Friday filing, Fat Joe's lawyers accuse Tyrone Blackburn, counsel for the rapper's former hypeman, of a motion with at least ten instances of cases cited for inapposite propositions or quotes found nowhere in them, "in all likelihood" from AI. They ask for "harsh sanctions." Blackburn, threatened with sanctions over AI errors in another case in June, did not immediately respond.
Who did it: HYBRID AI Hallucination Cases database
Hennepin County District Judge Christian Sande fined attorney David Lutz $5,000, struck his summary judgment briefs and referred him to the Lawyers Professional Responsibility Board. Both briefs cited a nonexistent case, Hallock v. Nelson. Lutz admitted using AI and failing to verify the citation through vLex Fastcase, and called the error unintentional.
Who did it: HYBRID AI Hallucination Cases database
Brazil's Regional Labour Court of the 2nd Region rejected a worker's appeal and fined the worker 5% of the claim's value for bad-faith litigation. The appeal had cited invented case summaries attributed to TST justices and to a nonexistent judge. The court dismissed the AI excuse, calling AI a mere tool and holding the worker responsible for the content.
Decode found that a user-made chatbot impersonating the late actor Sushant Singh Rajput, labelled "parody", had exchanged 3.7 million messages. It denied his death was suicide, named people it blamed, including his former girlfriend, and urged users to "seek justice". After Decode's inquiry, Meta removed the bot, saying it violated its AI Studio policy.
