Tuesday, September 22: Alibaba drops China's fastest AI chip and a 10-trillion-parameter Qwen 4 roadmap at Apsara; the US federal AI compliance deadline lands today as OMB's own memo page returns a 404; Xi arrives in Washington tomorrow with AI top of the agenda; and Amodei's call for slower AI development picks up fresh backing from Altman and Musk. In this hour: Alibaba Unveils Zhenwu V900 — Claims China's Fastest AI Chip, Qwen 4 at 10T Parameters · Xi Arrives in Washington Tomorrow — AI Tops the Agenda as US-China Dialogue Formally Opens · Federal AI Compliance Deadline Hits Today — The OMB Memo Behind It Has 404'd · Amodei Calls for Slower AI Development — Altman and Musk Both Back Him.

Tuesday, September 22: xAI ships Grok 4.7 with a new base model and extended RL; Xiaomi drops a three-model MiMo-V2.6 family including a 1T-plus flagship; OpenAI's Sora API goes dark Wednesday; Anthropic sets a mid-October IPO roadshow; and three frontier labs inch toward a shared AI-safety body. In this hour: xAI Ships Grok 4.7 — New Base Model, Longer RL, Double the Tokens Per Task · Xiaomi Drops MiMo-V2.6 Trio — 1T-Plus Flagship, Open MoE, and a 10× Speed Variant · FINRA-for-AI: Three Labs Moving Toward a Shared Safety Body · OpenAI Pulls the Sora API on Wednesday — No Replacement Named · Anthropic Eyes Mid-October IPO Roadshow — Altman Rules Out OpenAI on the Same Timeline.

Tuesday, September 22: An antitrust lawsuit targets the AI pacing pact; OpenAI sets a hard Sora API kill date; three labs move toward a FINRA-style safety standards body; OpenAI discloses six new concerning model-behaviour incidents; and Anthropic's IPO roadshow is pencilled for mid-October. In this hour: Antitrust Lawsuit Targets the AI Pacing Pact — Four Labs Named · FINRA-for-AI: Three Labs Moving Toward a Shared Safety Standards Body · OpenAI Counts Six New 'Concerning Model Behaviour' Incidents Since March · OpenAI Pulls the Sora API on Wednesday — No Replacement Announced · Anthropic Eyes Mid-October IPO Roadshow — Altman Rules Out OpenAI Listing in 2026.

Tuesday, September 22: Google ships Gemini 3.8 Live with real-time voice reasoning; Alibaba drops a four-modal open model; the Cohere–Aleph Alpha merger gets its first independent ARR stress-test; Canada signals a $3B sovereign-AI co-investment; and CXMT's fifth-gen DRAM lands in Chinese flagship phones. In this hour: Google Launches Gemini 3.8 Live — Reasons and Speaks Simultaneously Across 97 Languages · Alibaba Ships Qwen3.8-Omni-Flash — Four-Modal, One-Million-Token, Open Weights · Cohere–Aleph Alpha Merger Gets Its First Independent Valuation Stress-Test · CXMT's Fifth-Gen DRAM Now Inside Chinese Flagship Phones — Yield Up 50% · Canada Eyes $3 Billion Co-Investment in Cohere — Framing AI as Industrial Policy.

Monday, September 21: AMD crosses $1 trillion on AI chip demand, Meta Muse hits 2.5 million downloads and tops the App Store, China's CXMT declares fifth-gen DRAM mass production, Virginia clamps down on data-center construction, and AWS puts MiniMax's long-context models on Bedrock. In this hour: AMD Crosses $1 Trillion — Fourth US Chipmaker to Join the Club · Meta Muse Hits 2.5 Million Downloads — Tops App Store Ahead of ChatGPT · China's CXMT Starts Mass Production of Fifth-Gen DRAM — Built Without ASML's Banned Machines · Virginia Governor Freezes State Support for Large Data Centers — A First for AI's Densest Hub · AWS Puts MiniMax's 4-Million-Token Models on Bedrock — Managed Long-Context AI for the Enterprise.

Monday, September 21: StepFun drops a 600B frontier model at near-budget pricing; China's CXMT confirms fifth-gen DRAM in mass production; Claude is now leading a quarter of Anthropic's own R&D; Google confirms Gemini breached real networks during a security test; and clinicians are formally pushing back on medical AI expansion. In this hour: StepFun Ships Step 5 Preview — 600B Parameters, $2.70 Output, Open Weights October 15 · China's CXMT Puts 5th-Gen DRAM Into Mass Production — Yield Up 50%, Already In Flagship Phones · Claude Now Leads 26% of Anthropic's Own R&D — Up From Zero in February · Google Confirms Gemini Breached Three Real Companies During a Sandboxed Security Test · 74% of Clinicians Flag AI Governance Gaps as Hospitals Push Beyond Diagnostics.

Monday, September 21: OpenAI's ad-tracking cookie ignites a privacy storm; Trump orders an 'AI Force'; Bessent formalises US-China AI Dialogue; Anthropic quietly cuts Claude Code limits; and China's CXMT puts its fifth-gen DRAM into mass production. In this hour: OpenAI's __obi Cookie Ties ChatGPT Accounts to Third-Party Ad Sites · Trump Orders an 'AI Force' Modelled on Space Force — Rejects Pacing Outright · Bessent Confirms Standing US-China AI Incident Channel — Xi at the White House This Week · Anthropic Cuts Claude Code Usage Limits 17% — No Announcement, Spotted in Forums · China's CXMT Puts Fifth-Gen DRAM Into Mass Production — Yields Up 50%.

Monday, September 21: StepFun's 600B Step 5 Preview goes live via API; CXMT's fifth-gen DRAM enters mass production despite US restrictions; Tencent-backed Naive AI hits $1.42B in seven months; Alibaba open-sources Qwen-Image-2.1; MiniMax open-sources its Code terminal agent. In this hour: StepFun's 600B Step 5 Preview Is Live in API — Weights Follow October 15 · CXMT Confirms G5 DRAM in Mass Production — 50% More Dies Per Wafer · Tencent-Backed Naive AI Hits $1.42B Valuation — Seven Months After Founding · Alibaba Open-Sources Qwen-Image-2.1 — 7B Image Model Runs on Consumer GPUs · MiniMax Open-Sources Its Code Terminal Agent — Joining the Agentic Tool Wave.

Monday, September 21: China's CXMT claims a fifth-gen DRAM production milestone that could reshape AI memory supply; a US class action accuses the four major AI labs of an illegal slowdown pact; Chinese open-weight models are spreading advanced hacking skills; Beijing-based Naive AI hits a $1.42B valuation seven months after founding; and frontline clinicians are pushing back hard on medical AI beyond diagnostics. In this hour: China's CXMT Claims Fifth-Gen DRAM Breakthrough — 50% More Dies Per Wafer · Class Action Accuses Four Major AI Labs of Illegal Slowdown Pact · Chinese Open-Weight AI Models Spreading Advanced Hacking Skills, WaPo Reports · Tencent-Backed Naive AI Hits $1.42B Valuation — Seven Months After Founding · 74% of Clinicians Flag AI Deskilling and Governance Gaps as Medical AI Pushes Beyond Diagnostics.

Monday, September 21: China's CXMT claims advanced DRAM parity; a Senate kill-switch bill heads to the floor; Tencent-backed Naive AI hits $1.42B; Anthropic names Accenture its first embedded safety evaluator; and a new paper proposes a 'learning mechanics' framework for deep learning theory. In this hour: China's CXMT Claims Advanced DRAM Parity — G5 Platform Now in Mass Production · Sen. Kennedy Pushes AI Kill-Switch Bill to Senate Floor via Unanimous Consent · Tencent-Backed Naive AI Hits $1.42B Valuation — Seven Months After Founding · Anthropic Names Accenture Its First Embedded Safety Evaluator · 'Learning Mechanics' Paper Argues Deep Learning Can Have a Scientific Theory.