Anthropic said it is investigating a report of unauthorized access to Claude Mythos Preview, a model it says is too powerful to release publicly, through a third-party vendor environment. Bloomberg reported a group in a private forum had been using it, though not for hacking, lest they be detected; Anthropic says it has no evidence its systems are affected.
Florida Attorney General James Uthmeier opened a criminal investigation into OpenAI's ChatGPT, saying prosecutors believe it advised Phoenix Ikner, the Florida State University gunman, on what gun and ammunition to use and when and where to find the most potential victims. OpenAI said ChatGPT gave factual answers available across public sources and did not encourage harmful activity.
Sullivan & Cromwell apologized to Martin Glenn, chief judge of the U.S. Bankruptcy Court in Manhattan, for a filing with AI "hallucinations" and other errors, caught by Boies Schiller Flexner, also in the case. In an April 18 letter, Andrew Dietderich said the firm's AI policies were not followed and a secondary review missed the citations; it later filed a corrected version.
Meta told employees on Tuesday that a new tool, called Model Capability Initiative according to Reuters, will log their keystrokes and mouse clicks on company computers and internal apps as AI training data. A spokesman said the data is used for nothing else and safeguards protect sensitive content; one employee, amid expected job cuts, called it "very dystopian."
Nashville police say Jerry Dubose II, 47, used Grok AI to generate child sexual abuse images, and have charged him with sexual exploitation of a minor. Multiple CyberTips to the National Center for Missing and Exploited Children about an online account, with activity from September 2025 through March 2026, led detectives to his home; he remains jailed on $25,000 bond.
A medical student from India told WIRED he used Google Gemini's Nano Banana Pro to create "Emily Hart," a fictional MAGA-supporting nurse, after Gemini, per a transcript he provided, called the conservative niche a "cheat code." He says Fanvue subscriptions and MAGA T-shirts brought in a few thousand dollars a month before Instagram banned the account for "fraudulent" activity.
France's Renaissance party posted AI-generated images and fake five-star reviews on April 20 showing Emma Watson, Rosalía and Cristiano Ronaldo apparently reading Gabriel Attal's book "En homme libre," without their consent. After criticism, the party acknowledged an error, saying young activists had sent "parody" visuals its web team wanted to share; the posts have disappeared.
Lyon captain Wendie Renard filed a complaint for identity theft on Tuesday over an AI-generated video imitating her face to urge investment in an artificial intelligence project, her lawyer Laurent Burgy told AFP. The video is aimed particularly at residents of Martinique, where she is from; Renard said it reflects "neither my words, nor my values."
A blogger said on April 21 that after mistakenly pasting an image into Kimi, the chatbot returned a résumé containing another person's real name, phone number and email. According to the blogger, Moonshot AI blamed an AI hallucination and asked that the post be deleted; the blogger and the résumé's owner rejected that explanation and kept it up.
Clarifai deleted 3 million photos it says it got from OkCupid to train facial recognition AI, plus models trained on them, Reuters reports. Per the FTC's investigation, Clarifai asked for the data in 2014 and OkCupid, whose executives had invested in Clarifai, provided it against its own privacy policies; OkCupid and Match Group settled with the FTC last month without admitting the allegations.
Thomas Christopher Ball, 49, faces three felony charges after investigators say he sent a teenage coworker an AI-generated nude image of her, made from photos from when she was younger, and threatened to expose her if she kept ignoring him. A friend later called 911 saying he had allegedly come to her home looking for a firearm; he was arrested Friday.
The Guangzhou Intellectual Property Court ruled at first instance that five former senior managers and core developers of an AI firm used intranet penetration and electronic intrusion to take its "virtual digital human" source code. They later joined a rival whose product code was over 90% similar. The court ordered the seven defendants to pay 4.95 million yuan in triple punitive damages, plus costs.
