A Graphika report on five character platforms, Character.AI, Spicy Chat, Chub AI, CrushOn.AI and JanitorAI, found more than 10,000 chatbots labelled as sexualised, minor-presenting personas, with Chub AI hosting over 7,000 labelled sexualised minor female characters. It says creators use jailbreaks, local models and coded language to evade safeguards, alongside bots promoting eating disorders and extremism.
A day after launch, the Los Angeles Times' AI "Insights" tool answered Gustavo Arellano's column on the Ku Klux Klan in Anaheim by calling the group "'white Protestant culture' responding to societal changes rather than an explicitly hate-driven movement." The comments were removed; owner Patrick Soon-Shiong told CNN he had seen neither the column nor the AI response.
A leak obtained by SVT and analysed by OCCRP and partners shows a Tbilisi call-centre operation run by A.K. Group received $35.3 million from more than 6,179 "clients" between May 2022 and February 2025, according to its managers' records. Agents posed as financial advisers, and one kept a doctored British passport made with an AI-generated image.
Google told Australia's eSafety Commission it received 258 user reports of suspected AI-generated terrorist or violent extremist content made with Gemini, and 86 alleging AI-generated child abuse material, between April 2023 and February 2024. Google says the figures are global user reports, not confirmed policy violations; the regulator says Google used automated image-matching against abuse material but not extremist content.
A federal judge in California denied the preliminary injunction sought by Elon Musk, xAI and Shivon Zilis against OpenAI's for-profit pursuits, finding they "failed to meet their burden of proof for the extraordinary relief requested." Other parts of Musk's lawsuit, which alleges breach of contract and fiduciary duty and antitrust violations by OpenAI and Microsoft, can proceed.
Armenia's Personal Data Protection Agency says scammers used an AI-generated likeness of Prime Minister Nikol Pashinyan in a Russian-language video from an account called "Noticias Mundiales", promising "a stable income" of ֏1.8 million ($4,500) a month. Viewers were urged to register names, phone numbers and emails, with access supposedly open for only 24 hours.
A B.C. business analyst says callers posing as Amazon and his bank's fraud department urged him to transfer $10,000, even supplying a callback number with familiar on-hold music. He suspects his voice was cloned with AI in an earlier call to the bank; an assistant branch manager told him the bank sees the scam about twice a week.
Book authors including Sarah Silverman and Paul Tremblay, who allege OpenAI infringed their copyrights by training its models on their work, served former OpenAI researcher Alec Radford with a subpoena on February 25, a court filing shows. OpenAI maintains the training is fair use; a magistrate judge separately ruled Dario Amodei must sit for questioning.
