Overview
- An EBU/BBC-coordinated review of more than 3,000 answers across 18 countries found 45% of AI assistant responses had at least one major problem, including serious sourcing issues in 31% and major accuracy errors in 20%, with Gemini scoring worst at 76%.
- The study team released a News Integrity in AI Assistants Toolkit and urged EU and national authorities to enforce existing rules on authenticity, digital services and media plurality, signaling plans for rolling independent monitoring.
- OpenAI said roughly 0.15% of weekly users show explicit suicidal intent in chats—over one million conversations given more than 800 million weekly users—and reported GPT-5 achieved 91% compliance with its suicide-response guidelines versus 77% previously after input from over 170 clinicians.
- SoftBank approved a further $22.5 billion investment contingent on OpenAI reorganizing into a public benefit corporation to enable an IPO, bringing its planned total to $30 billion, with reports noting the tranche could shrink if the restructuring slips beyond year-end.
- Financial Times reporting describes Sam Altman negotiating large multi‑year chip and infrastructure commitments directly with suppliers such as NVIDIA, AMD and Oracle with limited disclosed terms, while The Wall Street Journal highlighted a $4.7 billion equity‑method loss in Microsoft’s FY2025 results that investors suspect may include OpenAI‑related losses ahead of Wednesday’s earnings.