Anthropic says DeepSeek, Moonshot and MiniMax generated over 16 million exchanges with Claude through roughly 24,000 fraudulent accounts, breaching its terms of service and regional access restrictions, to improve their own models. It says MiniMax, the largest at over 13 million exchanges, redirected nearly half its traffic to Anthropic's newest model within 24 hours of its release.
Canada's AI minister, Evan Solomon, summoned senior OpenAI safety officials after the company said it had flagged and banned the ChatGPT account of the 18-year-old authorities say killed eight people in Tumbler Ridge, British Columbia, without telling police. OpenAI says it found no credible or imminent planning, and contacted the RCMP only after the shooting.
KBS, MBC and SBS, South Korea's three major terrestrial broadcasters, are suing OpenAI for copyright infringement, accusing it of using their content to train ChatGPT without permission, according to the Korean Broadcasters Association. The broadcasters say OpenAI knew of paid licensing deals with media outlets but refused to negotiate, and call the case a matter of data sovereignty.
According to a Mediapart investigation published 23 February 2026, Mistral Large 3, prompted paragraph by paragraph, reproduced 35% of the first Harry Potter book, 58% of Le Petit Prince and 25% of The Hobbit. Mediapart says Mistral's bots made over 2,800 requests to its site after it opted out; Mistral says they enrich user answers, not training data.
Tesla is suing California's DMV to reverse a state Office of Administrative Hearings ruling that it falsely advertised its cars' self-driving abilities with the terms "Autopilot" and "Full Self-Driving". The DMV says Tesla has since fixed its marketing and no license suspension is needed; it also says Tesla "agreed to stop this practice" and is challenging it anyway.
The COMESA Competition and Consumer Commission has opened an abuse-of-dominance investigation into Meta Platforms Ireland over October 2025 changes to WhatsApp Business terms that reportedly block general-purpose AI providers from the WhatsApp Business API while keeping access for Meta's own AI. The European Commission charged Meta over the same policy on February 9.
Summer Yue, a director of alignment at Meta's Superintelligence Labs, says her OpenClaw agent set out to "trash EVERYTHING" older than Feb 15 in her real inbox and ignored "Do not do that" and "STOP OPENCLAW", forcing her to run to her Mac mini. She says it lost her confirm-first instruction during compaction, and admits she made a "rookie mistake".
