📰 Daily Briefing · 2026-09-17 Thursday

🌐 Politics · Geopolitics

• No major geopolitical developments today; the Fed decision dominated the global agenda. World attention converged on Washington: the first rate hike in over three years, an upward-revised dot plot, and Chair Warsh's hawkish press conference formed the single global storyline of the past 24 hours.
• China platform governance: sweeping new rules for financial content — Douyin, Xiaohongshu and other platforms rolled out fresh financial-content policies, with simultaneous purges of violating accounts, tightening the regulatory framework around finance-focused self-media. ↳ Takeaway: Platforms are self-regulating ahead of formal regulation — the "financial credentials" threshold is being hard-coded into recommendation systems.　·　CLS
• Former Vanke chairman Yu Liang placed under consumption restrictions — Yu Liang has been hit with a high-spending ban, another sign that personal liability for executives is surfacing during the prolonged property downturn. ↳ Takeaway: The property unwind has moved from corporate balance sheets to named individuals — a new, more personal phase of the cleanup.　·　CLS

💰 Finance · Markets

Snapshot (US close 9/16 ET · A-share close 9/16): S&P 500 7551.81 (-0.45%) · Nasdaq 25978.42 (-0.01%) · Dow 51461.90 (-1.21%) · Shanghai Composite 3891.60 (+0.71%) · CSI 300 4480.27 (+0.68%) · USDCNY ≈6.698 · Oil / US 10Y / Gold: not available

• First Fed hike in over three years: 25bp to 3.75%-4%, dot plot signals one more this year — The FOMC voted unanimously to lift the federal funds target range to 3.75%-4%, the first increase since July 2023; 18 officials now project rates at 4%-4.25% by end-2026. The statement called inflation still elevated and said the move helps return it promptly to the 2% goal. ↳ Takeaway: The easing cycle that began in September 2024 is officially over — "inflation is too high and has persisted for too long" is about as unambiguous as Fed language gets.　·　CLS
• Chair Warsh goes hawkish: inflation too high for too long, underlying trend not clearly improving — Warsh said the economy is strong enough to absorb further hikes, and the 10-year Treasury yield climbed back above 5%; markets read it as "higher for longer." ↳ Takeaway: Unanimous vote + revised dots + hawkish rhetoric — a triple signal that this tightening is a trend, not a one-off.　·　CLS
• US close: hawkish Fed weighs on sentiment, all three indices lower — The Dow fell 1.21% to lead declines; retreating oil dragged Chevron down 2.9% and ExxonMobil 3.5%. Deutsche Bank and KPMG economists confirmed the tightening cycle has begun. ↳ Takeaway: The bond market's verdict is the interesting part — restored credibility made the hawkish outcome bond-positive, giving stocks and bonds opposite grades for the same hike.　·　CLS, Sina Finance
• US mortgage rates break above 7% — The hike pushed mortgage rates past the 7% mark, ratcheting up funding costs for households. ↳ Takeaway: 7% is the sensitive threshold for US housing — last time it broke through, existing-home sales weakened immediately. Watch whether the data cooperates this time.　·　CLS
• A-shares chart their own course: STAR 50 up over 4%, compute hardware and semiconductors lead, 4,100+ stocks gain — The ChiNext recovered from a weak open to close up nearly 2%, multiple PCB stocks hit record highs, and a newly listed compute ETF surged on debut — growth/tech dominated a broad rally on rising volume. ↳ Takeaway: A-shares running their own compute narrative right as the Fed tightens — the divergence in pricing logic between the two markets is the real story, not the index moves.　·　CLS

🚀 Tech · AI

• Korea launches $1B national physical-AI R&D program — The "Gyeongnam–Jeonbuk AX" project runs 2026-2030 with a 1.41 trillion won (~$1.01B) budget: Gyeongnam focuses on ultra-precise human-robot collaborative AI control, Jeonbuk on multi-robot factory coordination software. Korea previously targeted commercial humanoids for 10 key industries by 2028. ↳ Takeaway: Physical AI has graduated from paper concept to national budget line — once state funding enters, foundation models for robotics become an infrastructure race.　·　CLS
• Novo Nordisk partners with Anthropic to accelerate drug discovery with Claude — The pharma giant will use Claude across its R&D workflow, adding a heavyweight case to AI-driven drug development. ↳ Takeaway: The insulin maker is buying "intelligence" — AI monetization is shifting from selling subscriptions to embedding in industrial workflows.　·　CLS
• Jensen Huang doubles down: no new AI laws needed, safety and speed can coexist — Speaking at a Salesforce event, the NVIDIA CEO argued market forces will push companies to innovate safely without new regulation. ↳ Takeaway: The "legislate vs. peer-review" split didn't converge even on Fed day, but big labs' stance is clear: run first, tie shoelaces while running.　·　Sina Finance

📚 24h Blogs · Papers

• Overseas + Chinese blogs (Lilian Weng / Anthropic Engineering / Cognition / Manus / Baoyu / Qwen / DeepSeek / Jiqizhixin): no updates within 24h today. The 9-16 timestamp on Anthropic Engineering's listing page turned out to be a site-config update; the latest post is still April's Claude Code quality-reports postmortem. Cognition's 9-15 item (Cognition × AWS) was covered in yesterday's briefing.
• arXiv cs.AI picks (Sept 15 batch):
  • Agentic Societies Need a Social Harness — Argues multi-agent societies need not stronger individual models but a "social constraint layer," echoing Lilian Weng's earlier harness series.
  • When Should LLMs Abstain? Chain-of-Self-Questioning for Selective Risk Control — Has models interrogate themselves before answering to decide whether to abstain, turning "knowing what you don't know" into a calibratable risk-control mechanism.
  • ScienceBuddy: Recursive-in-Recursive Self-Improvement for Interactive Scientific Agents — A recursive self-improvement framework for interactive research agents; self-improvement is moving from slogan to architecture.
  • ↳ Takeaway: Read together, the three trace one clear line — the next step for agents isn't "smarter," it's "better-behaved, more aware of their limits, and capable of iterating on themselves."

🐙 GitHub Trending (new repos · 3-day window)

• theoephraim/awesome-cloudflare-selfhosted ⭐237 — A curated list of open-source apps that replace SaaS products, running in your own Cloudflare account; "self-hosting on Cloudflare" is becoming its own category.
• opencoredev/bg0 ⭐205 — Fully local, in-browser background removal that never uploads your images — a benchmark for privacy-first tooling.
• mpociot/claude-siri-ai ⭐188 — An experimental macOS 27 App Intents model-delegation provider that routes Siri requests to Claude Code — grassroots exploration of system-level AI takeover.
• KillaBoi/BrokenPipe ⭐175 — A Steam client service local privilege escalation vulnerability (PoC), a fresh exposure in gaming-platform security.
• Marcos66236/github-stars-history ⭐227 — Tracks and visualizes star history and growth velocity for any GitHub repo; open-source growth analytics as a tool.
• Frankweb33/flybrain-robot-bridge ⭐153 — A Drosophila-vision-inspired camera-to-robot bridge with optical flow + IMU feedback over a mock neural backend — a toy-grade take on biologically inspired robotics.
• ↳ Takeaway: claude-siri-ai is the one to watch — when hobbyists start wiring Claude into OS-level entry points, "agent as the interface" stops being just a big-lab narrative. Filtered ~5 low-value repos (ai-sucks-butt, a pure meme with 1,662 stars, a hacker-forum directory, and no-description projects); page-mascot, video-generator-client, ToolReplay and others were covered in yesterday's briefing and aren't repeated.

---

Market data via Sina Finance API (US close 9/16 ET); oil, US 10Y and gold were not available today and are marked as such. Sources: CLS (Cailian Press), Sina Finance, arXiv.