US-based security firm Enkrypt AI reported that Mistral's Pixtral-Large (25.02) and Pixtral-12b, fed prompts hidden in images, gave detailed suggestions for scripts to convince a minor to meet for sexual activities, and were 60 times more prone to generate child sexual exploitation material than GPT-4o and Claude 3.7 Sonnet. Mistral said it has "a zero tolerance policy on child safety".
Fraudsters who earlier used Baykar chairman Selçuk Bayraktar's face and voice in fake stock-tip videos have turned to Güler Sabancı. The former Sabancı Holding chair warned that AI-made fake videos using her name, image and the holding's logo are circulating as "investment advice" on Facebook, Instagram, YouTube, WhatsApp and Telegram, and said legal proceedings had been started.
Morphisec says criminals are promoting fake AI video-generation sites through Facebook groups, one post drawing over 62,000 views. Users who upload images to get a "processed" video instead download a ZIP containing Noodlophile Stealer, a previously undocumented infostealer that harvests browser credentials and crypto wallets, sometimes bundled with the XWorm remote access trojan.
Microsoft says a network it calls Storm-2139 used stolen Azure customer API keys and custom tools to bypass safeguards in Azure OpenAI Service, creating thousands of abusive images of celebrities, women and people of color. Its civil complaint in the Eastern District of Virginia names four defendants, in Iran, England, Hong Kong and Vietnam, and alleges 10 others used the tools.
Morpheus Research, which holds short positions in Mercurity Fintech, alleges the 11-person Nasdaq company's AI-server liquid-cooling venture is an empty promotion: its unnamed partner is ZJK Industrial, a nuts-and-bolts maker, and the venture's Shenzhen subsidiary address did not exist, according to the report. It says the announcements fuelled a 418% share rise since November 1, 2024.
AI-generated clips of the late Umm Kulthum singing rapper Wegz's song "El Bakht", with the singer's synthesised voice laid over footage from a concert, stirred controversy on Egyptian social media. The family threatened to go to court to protect the singer's image; a grandniece called the fabricated videos "provocative" and said they had saddened the family.
Scammers used AI to publish counterfeit versions of a Holocaust survivor's memoir within days of its release, under fake author names including "Jude Williams" on Amazon, which the book's co-author said carried "a blatant anti-Semitic slant". The survivor said the copies have been taken down; Goodreads reportedly argued the reworked text was not copyright infringement.
NSW police minister Yasmin Catley revealed on 8 April 2025 that NSW Police had stopped using Cognitec's PhotoTrac facial recognition in its suspect identification system, without giving a reason. Police had still been running the original 2011 algorithm; the US National Institute of Standards and Technology found the latest version misidentified West African people almost seven times more often than Europeans.
