Skip to content
Source · Daily Brief

AI Daily Brief — 15 September 2025

Apple’s day. iOS 26, macOS Tahoe 26, watchOS 26, tvOS 26, visionOS 26 all shipped together. Anthropic Usage Policy hardened. Samsung opened AI Forum. China shipped neuromorphic LLM on domestic chips.

Top stories

  • iOS 26 ships with Apple Intelligence. iOS 26, iPadOS 26, watchOS 26, tvOS 26, visionOS 26 to public. Biggest Apple Intelligence update of year: on-device Live Translation in Messages/FaceTime/Phone + via AirPods. Visual Intelligence lets users screenshot anything and ask ChatGPT. Intelligent Shortcuts actions, Workout Buddy, Genmoji merging existing emoji, new Liquid Glass design across all platforms. via Apple
  • macOS Tahoe 26 released — biggest Mac redesign since 2013. Liquid Glass interface (translucent, reflective), native Phone app on Mac, Spotlight Actions, Clipboard History, tighter Apple Intelligence integration. Tahoe final macOS to support Intel Macs. via MacRumors
  • Anthropic Usage Policy update takes effect. Revised policy live with dedicated section on prohibited malicious computer/network/infrastructure compromise — written explicitly to address new attack surface from Claude Code and Computer Use. Tightened restrictions on political deception/voter targeting while preserving political research. Clarified high-risk-use-case requirements apply only when outputs consumer-facing. via Anthropic
  • Samsung AI Forum 2025 opens. Ninth annual forum (Sep 15-16) in Korea. Day 1: “Vertical AI Strategies and Future of Semiconductor Industry”. Yoshua Bengio (Université de Montréal) keynote on “Scientist AI” — system designed to give truthful, evidence-grounded answers — and AI for materials discovery. Stanford’s Stefano Ermon (co-founder of Inception, diffusion-language pioneer) also spoke. via Samsung
  • SpikingBrain 1.0 — brain-inspired LLM on domestic MetaX GPUs. Chinese Academy of Sciences (Institute of Automation, Beijing) released neuromorphic LLM using “spiking computation” (only necessary neurons fire). 7B variant ran ~100x faster than comparable transformer on 4M-token prompt. Also released 76B version. Training and inference run entirely on Shanghai-based MetaX GPUs — pointed demonstration of non-NVIDIA Chinese AI stack. via SCMP

Who shipped

Apple shipped iOS 26 + Apple Intelligence wave. Anthropic shipped policy update. Samsung shipped AI Forum. CAS shipped SpikingBrain.

Open-source pulse

Qwen3-Next-80B-A3B ecosystem widens — week-after-launch attention. Alibaba’s Qwen team’s new architecture (80B-parameter MoE that activates just 3B per token, hybrid Gated DeltaNet + Gated Attention, 256K context, ~10x throughput vs Qwen3-32B at long context) drew sustained coverage. Both Instruct and Thinking open-weight on HF; Thinking benchmarks above Gemini 2.5 Flash Thinking.

Money, infra & hardware

Apple debuts iPhone 17 lineup (on sale Sep 19; Apple Intelligence-first marketing) — all four models (17, 17 Air, 17 Pro, 17 Pro Max) sold as Apple Intelligence flagships, expanding install base capable of running on-device generative features.

Quiet corners

Kazakhstan moves on dedicated AI ministry — President Tokayev’s state-of-the-nation address “Kazakhstan in Era of Artificial Intelligence” laid out plans for Ministry of AI and Digital Development headed by Deputy PM, plus Digital Code and Crypto Asset Fund. Tesla Optimus + xAI Grok integration story spreads. Mistral × ASML strategic partnership in market consciousness.

By the numbers

  • iOS 26 / Tahoe 26 / watchOS 26 / tvOS 26 / visionOS 26 — all shipped together
  • 2013 — last Mac redesign before Tahoe
  • 9th — annual Samsung AI Forum
  • 7B / 76B / ~100x / 4M tokens — SpikingBrain sizes / speedup / prompt
  • MetaX — Shanghai-based GPU running SpikingBrain

Compiled by AI Feed’s editor from verified web sources for 15 September 2025.