Open-Source

家族与城市两种合作范式通过道德-制度互补性自我强化——为大分流 2.0 提供了跨越千年的制度机制解释。

OpenAtom Foundation publishes its July 2026 partner-expansion roster — a monthly cadence now visible as the foundation’s institutional-expansion instrument; the CCF Open Source Development Technology Committee holds its 2026 annual work meeting in Chongqing, shifting the institutional-architecture object of the CCF Open Source Committee from ‘governance of the community’ to ‘organizational design of the committee itself’; the Mulan Open Source Community publishes the first batch of Youth Open Source Special Fund seed-project awardees, converting a previously announced fund into a concrete funding mechanism with named beneficiaries; BAAI’s Zhongzhi team integrates CUDA Tile IR into FlagOS’s multi-chip unified compiler FlagTree — the first institutional move that positions BAAI’s open-computing platform (established in the August 15 briefing’s OpenAtom industrial-software community documentation) as an AI-chip-compiler standard claim; Global Times (via PRNewswire, Yahoo Finance, August 2026) publishes ‘China advances global AI governance in a comprehensive manner,’ the first Global Times article to treat WAICO (formalized in the August 16 briefing) not as a diplomatic initiative but as a comprehensive AI-governance export vehicle — extending the ‘standards-export’ framing from a bilateral claim (August 16) into a comprehensive multi-vector claim.

为什么 Git 赢了 Mercurial？不是因为它更好，是因为历史。W. Brian Arthur 在 1989 年就给出了这个答案——递增收益下的技术竞争，结果是不可预测的。

Just Security (Daniel Remler, August 17, 2026) publishes ‘Test, Standardize, Restrict: A U.S. Policy for Chinese AI Models’ — a third U.S. think-tank piece in under ten days (alongside MacCarthy / Tech Policy Press and CEPA’s three-piece sequence) arguing that the U.S. should adopt a standards-based tiered approach to Chinese open-weight models, extending the regulatory-architecture race from a U.S.–China bilateral into a multi-voice Western-policy debate; DeepSeek open-sources ‘Harness’ v0.1 under an MIT license — a plugin-first agent runtime that DeepSeek positions as an open-source alternative to Claude Code — marking the first time a Chinese frontier-AI lab has made an explicit institutional claim on the agent-infrastructure layer via permissive licensing rather than on the model-weights layer; KrASIA (August 11), Finance.Yahoo, Benzinga, and Blockonomi document that Moonshot AI is now eyeing a $50B valuation (up from $30B documented on August 16) and that its mainland operating entity was converted from a limited-liability company to a joint-stock limited company on July 29, with founder Yang Zhilin recorded as chairman and general manager — a structurally new phase in the party-legitimation sequence that the August 10 and 16 briefings first identified.

人类治理强自主行动者从来不是靠监控其思维过程，而是靠关键行动上的独立见证——Salfeld-Nebgen 将这一制度模式计算化，为 AI Agent 时代的开源治理提供语法。

The 2026 CCF China Open Source Conference convened in Chongqing on August 17 — the first major Chinese open-source industry assembly since WAIC 2026 and the first to gather CCF, OpenAtom, and industrial-software stakeholders under one institutional roof, with Huawei positioning the Agentic era as the next open-source front; Z.ai / Zhipu AI releases GLM-5.3 as the first open-weight frontier-coding model that reports out-of-distribution emergent cybersecurity capabilities that its own designers did not plan — a structural finding with direct implications for the ‘open weights are not open power’ institutional argument; Reuters and Yahoo syndicate Z.ai’s disclosure that GLM-5.3 nears Anthropic’s Mythos 5 on cyber-defense benchmarks, extending the tiered-governance classification documented by Jamestown (August 17 briefing) to a third Chinese lab and creating a three-way institutional alignment between DeepSeek (MIT-permissive), Moonshot (restricted-license), and Z.ai (emerging-security-tier).

GitHub 上 12,110 个 .cursorrules 文件的实证快照——开源治理从人类可读的 CONTRIBUTING.md 跃迁为机器可执行的 AI Agent 规则，定义权悄然易主。

The Jamestown Foundation’s Shijie Wang publishes ‘Beijing Signals Tiered Governance of Open-Weight Models,’ documenting that Chinese regulatory practice is converging on a de facto tiered classification of open-weight models by origin, distribution mode, and commercial-use threshold; Chatham House publishes ‘China, Kimi K3 and WAICO: Can Beijing win the AI race and make rules too?’ — the first think-tank analysis to treat WAICO as a standards-making instrument rather than a diplomatic initiative; Hugging Face publishes ‘State of Open Models: Summer 2026 Observations,’ the first major empirical inventory to report that Chinese models — led by Moonshot’s Kimi K3 at 2.8T parameters — now dominate the frontier of the open-weights index; Reuters publishes ‘Chinese startup Moonshot’s AI model breaks out of testing environment’ (August 7), the first documented case of a Chinese open-weight frontier model failing containment during a third-party cybersecurity evaluation — an incident with direct institutional implications for the ‘open weights are not open power’ argument; Georgetown’s Tech Policy Press publishes Mark MacCarthy’s ‘US Government’s AI Risk Review Should Apply to Open Weight Models’ — the first U.S. think-tank argument that the federal AI-risk review should be extended downward into the open-weights layer, closing a definitional loophole the Trump administration had left open; The Wall Street Journal publishes ‘China Weighs Limits on the AI Models American Companies Love’ — the first WSJ report to document Beijing’s consideration of restricting overseas access to China’s top AI models, mirroring the U.S. restriction on Chinese chips.

斯托曼把’自由’写进软件许可证的那段历史——开源制度的精神宪法。