Current developments in generative AI, software engineering, programming, and digital products, with links to original sources.
OpenAI Disrupts Russia-Based Fake Think Tank Operation ── Mystery 'Ox Alpha' Takes Top Usage Share, Lawyer Fined Over Hallucinated Citations ‖ AI News for August 25
SSRF Vulnerability in Google's Official MCP Toolbox Fixed in 8 Days — a 2.8-Trillion-Parameter MoE on a 4070 Ti, and the “Amounts Not in the Source Text” That LLMs Invent
Andrew Ng Refocuses on 'AI Engineering' — Mozilla Report Shows Chinese Models Dominating the Top Open Models, ChatGPT Plus Brings Back 5-Hour Limits
Apodex 1.1 Takes on Complex Work with Asynchronous Agent Teams — Reports of Civilian Casualties from AI-Guided Drones, and 80% of Developers Find AI Coding "Addictive"
SEC Probe Hits AI Hedge Fund — Lambda in Talks to Raise $3 Billion at a $12 Billion-Plus Valuation, and DeepSeek's Agent "Escapes" Its Working Folder ‖ AI News for August 25
NVIDIA's "Vera Rubin" Delivers 4x the Closest Competitor in Agent Inference — Alibaba's Wan3.0 Generates 30-Second Videos for $6, JetBrains Unveils Qwen3.6 27B Local AI ‖ August 25 AI News
Thomson Reuters Unveils $40M In-House LLM "Thomson" — Taiwan Prosecutors Indict Nvidia Executive, METR Analysis Reveals the "Unevenness" of AI Acceleration
Xiaomi Challenges Qualcomm With Its Own AI Chip “Xring O3” — Cerebras’ CS-4 Doubles Performance on the Same Chip, and UK AI Investment Is Back at Dot-Com Era Levels ‖ AI News for August 24
Why Children Learn Language from Far Less Data Than AI — 30B-Class Open Models Heat Up, Nvidia in Talks to Invest in Perplexity at Over $30B Valuation ‖ AI News, August 24
Hugging Face Reportedly Up for Sale in a $13B-Scale Deal — Xiaomi's AI Cube Prototype, a Claude Outage, and a 150M Model Designed Backward from CPU Inference ‖ AI News, August 24
