Vu 等（2026, arXiv 2606.20669）提出 Agent Behavior Mining 与「invisible autonomy risk」概念——把流程挖掘引入 AI agent 治理，让 agent 的推理、工具、token 成本转为可观测的过程日志。这是 Coase「企业为什么存在」命题在 AI 时代的当代表述：当企业内部隐性协调成本被 agent 非确定性放大时，企业边界与俱乐部章程都必须被重新界定。开源的信任基础从「代码可见」升级为「行为可见」。
“MIIT’s office issues the AI + Software Special Action Implementation Plan on September 11, 2026 — the seventh documented surface on the Chinese open-source institutional-architecture question that this series’ prior twenty-four briefings have been tracking — making five institutionally new moves that extend the September 6 briefing’s MIIT AI SME Startup Support Plan (31号) on the same unsolved object: (1) the plan formally directs the construction of an ‘Agent Software Application Store’ (智能体软件应用商店) and a ‘Skills Resource Library’ (Skills 资源库), a specific state-directed institutional apparatus for agent-software distribution and skill-pack distribution — the same distribution-surface move this series documented at the Android-app-store layer in the September 3 CAC Qinglang Phase 2 briefing, now operationalized at the agent-software layer; (2) the plan explicitly names ‘independent open-source license application and compliance guidance’ (自主开源许可协议应用与合规指引) as a supported direction, the first administrative-document use of Mulan-family-license promotion as a national industrial-policy instrument; (3) the plan introduces a quality-open-source-project selection-and-cultivation mechanism with open-source project first-release at community (社区首发), an institutional-design move that pairs a state-directed project-selection mechanism with a community-first-release operationalization; (4) the plan explicitly directs that open-source contributions be incorporated into research-evaluation and talent-certification systems at universities, research institutes, and enterprises — extending the September 6 briefing’s star-count-as-administrative-metric move (which placed GitHub-style star counts alongside invention patents and highly-cited papers as one of four national-level evaluation indicators) into a broader research-evaluation-incorporation move at the higher-education and research-institute layer; (5) the plan pairs the open-source moves with a 2028 target of 5+ quality open-source projects (5个以上优质开源项目) and a 2030 target of multiple open-source communities with international influence — the first time a national industrial-policy document attaches quantitative open-source-project targets to the industrial-policy horizon. Together, the five moves document a structurally new institutional-form object: the state-directed-app-store / independent-license-compliance / community-first-release / research-evaluation-incorporation / quantitative-project-target pairing is the sharpest documented formalization of open-source as a state-directed industrial-policy instrument at the national-industrial-policy surface, and it pairs institutionally with this series’ prior briefings’ documentation of open-source at the administrative-control layer (CAC, September 3), the party-theoretical-flagship layer (Chen Kaihua, Qiushi, September 11), the Western-foundation-governance-seat layer (PyTorch Foundation, September 10), and the administrative-industrial-policy layer (AI SME Startup Support Plan 31号, September 6).”
奥斯特罗姆 1990 年用数十年跨文化田野研究证明：没有国家也没有市场，人依然能自组织地共同管理公共资源。她给出的八条设计原则——边界清晰、集体选择、监督、分级制裁、冲突解决、组织权认可、嵌套治理——是开源社区治理的直接理论来源。这是 2009 年诺贝尔经济学奖首位女性得主的标志性著作，也是开源制度分析工具箱里最关键的一根承重柱。
“OSPP (Open-Source-Summer / 开源之夏, the Chinese government-backed Summer-of-Code-type program initiated by the Institute of Software of the Chinese Academy of Sciences) announces its 2026 Phase 2.0 redesign with five formal institutional-design changes — (1) the funding gate is now restricted to foundational-software and RISC-V ecosystem projects, (2) enterprises are opened as first-time participants alongside open-source communities (previously community-only), (3) a new Pioneer Mode (领航模式) inverts the problem-definition authority from organization to student under a committee-invitation gating mechanism, (4) all fixed deadlines are removed in favor of year-round rolling operation, and (5) the compensation-scale is priced at 3,500–5,000 yuan for the classic mode and 6,000–10,000 yuan for the Pioneer Mode across a 2–6 month development cycle; alongside the OSPP 2.0 anchor, this briefing documents for the first time a three-tier competition operationalization across three institutionally parallel organizer scales — the community-scale Hack-Astron × openKylin Skill Development Challenge (500 yuan, September 9 briefing, scoring open-source-contribution-norms alongside technical criteria), the enterprise-scale Alibaba Cloud Qwen3.8-27B Experience Contest (10,000 yuan cash prize pool, operating as a user-behavior-elicitation mechanism), and the municipal-government-scale Zhengzhou 2026 Agent Innovation Application Competition (Zhengzhou City Industry and Information Technology Bureau + Zhengdong New District, three tracks including a directed-specific track where enterprises pose problems and the competition judges on enterprise-accepted solutions, finals September 18–22); the two-axis-parallel-operationalization pattern — community-scale-and-government-scale operationalization (OSPP 2.0) plus competition-scale-and-organizer-scale operationalization (three-tier competition) — is a structurally new institutional-form object on the Chinese open-source talent-pipeline-and-technical-priority-agenda surface that prior cycles had not yet observed together.”
Hora/Robbes/Zacchiroli (2026, arXiv 2609.07542) 首次对主流开源项目 AI 贡献政策做横向普查——281 份政策 + 92 份专用文件时间序列，得出 83.3% 允许 / 67.3% 要求人类参与 / 48.8% 要求披露 / 半份已修订四数据点。同一批作者 4 个月内样本量从 118 扩到 281（+138%），「AI 贡献政策」作为独立治理制品已跨过临界点。Hayek 自发秩序的教科书级实证 + Coase 交易成本在开源治理层的直接应用。
“Four institutionally distinct movements on four different surfaces — one organizational (Huawei’s open-source ecosystem lead Huang Zhipeng departs after 12 years with no named successor, on the same week Huawei’s PyTorch Foundation Platinum entry is being formalized in a Western neutral foundation), one capital-market (DeepSeek engages CITIC Securities as STAR Market pre-IPO tutor, marking the formal institutional transition from private frontier AI lab to public-listed entity under Chinese capital-market regulation), one cross-border-capital (Moonshot AI considering dual Hong Kong + Shanghai STAR Market listings after a confidential Hong Kong IPO filing, with the SCMP explicitly noting that Hong Kong-listed AI stocks are underperforming and that the STAR Market’s relaxed profitability requirements make it a natural second venue), one WeChat-monitor diagnostic (早安胡同’s argument that 国产系统 should be framed as ‘institutional products’ rather than ’technical products’, with a specific four-characteristics framework that reads 麒麟 / 统信 / 鸿蒙 as institutional-positioned operating systems, not just Linux derivatives). The four events together document a structurally new institutional pattern: on the Chinese open-source layer, the same week that Huawei’s internal open-source leadership continuity is being questioned (Huang Zhipeng’s departure) and DeepSeek’s and Moonshot’s capital-market formalization is being operationalized (STAR Market pre-IPO tutoring, dual-listing exploration), the WeChat-monitor surface explicitly formulates the domestic-OS question as an institutional-product question — the same question this series’ August and September briefings have been documenting at different institutional surfaces, now operationalized at the WeChat-monitor-popular-commentary layer with a specific four-characteristics framework.”
Kurtz & Krawiecka (2026, arXiv 2604.06148) 提出跨企业与地缘边界的 AI 系统机器身份治理分类法（MIGT）——企业环境中机器身份与人类身份的比例超过 80:1，单个未治理 Agent 在 2024 年 CrowdStrike 事件中造成 $5.4–10B 损失。四项原创贡献（AIRT 风险分类 + MIGT 六域框架 + 国家行为体威胁模型 + 跨辖区监管对齐结构），首次把「制度基础设施第五层：Agent 信任基础设施」从命题升级为可操作的治理框架。同时给出 EU/US/中国监管义务不可调和冲突的制度环境实证——「评价体系不可通约性」在监管层的复现。
“Three institutional objects on three different surfaces — one ideological (Qiushi, the Party’s theoretical flagship), one diagnostic (Zhishashe, an independent institutional-economics analysis piece), one competitive (a WeChat-monitor Xinchuang-industry half-year scorecard). First, Chen Kaihua’s August 31 article in Qiushi (2026/17), a State Council-affiliated policy-researcher profiled as an invited professor at UCAS, formally places open source at the center of a national scientific-research reform agenda — for the first time, on a Chinese party-theoretical surface, listing 开源社区 (open-source community) alongside data platforms and AI research agents as a formal 科研主体 (research subject), and proposing ‘a national-level open-source community for AI-driven research’ with an explicit mechanism combining ‘code contribution and compute incentive’ (代码贡献与算力激励) — a structurally new institutional formulation on the same research-paradigm question that the September 9 briefing’s CCTV codification first broadcast to a general audience. Second, Qinghe, editor-in-chief of Zhishashe (智本社), publishes a ~7000-word institutional-economics comparative analysis — ‘中国大厂：两边通吃’ (Chinese Big Cos: Eats Both Sides) — comparing commercial-credit behavior across 5,382 A-share non-financial firms versus 1,523 U.S. non-financial firms, documenting a 5.33-trillion-yuan net payable-receivable surplus on the Chinese side (8.4% of revenue) against a 0.14-trillion-dollar net deficit on the U.S. side (−1.2% of revenue), and reading Chinese large manufacturers as institutional ‘shadow banks’ (影子银行) with interest-income-to-interest-expense coverage rising from 38% to 97% over 2017-2025 — an independent diagnostic that pairs directly with this series’ August 26 Heilan ‘going-global’ institutional critique, applying the same institutional-economics frame to a different institutional surface (domestic commercial credit, not export). Third, Xinchuang-Lutoushe (信创露透社) publishes a September 1 half-year scorecard for 36 Xinchuang vendors across 8 segments — the sharpest institutional-economics finding being the divergence between the hardware layer (Cambricon 半年净赚23亿 turning profitable at 2.3B yuan, Muxi turning profitable at 0.6B yuan, Hygon at 9.1B revenue 12.16x the next-largest domestic-CPU rival) and the integration-vendor layer (Taiji dropping its profit 281x to a 350M yuan loss on a 72% market-cap decline from its April 2023 peak) — an institutional-form observation about where the domestic-substitution process is producing cash flow versus where it is producing debt and loss. One perspective, not a verdict.”
Grgic, Maksimovic & Laskov (2026, arXiv 2609.05380, ARES 2026 Workshops) 提出软件供应链攻击的四阶段传播模型（结构暴露 → 漏洞类存在 → 代码可达性 → 污点路径分析），用 Syft / Trivy / cdxgen / Grype 四个主流开源 SBOM 工具对 Log4j 漏洞做实证——结果显示：开源 SBOM 工具系统性只覆盖 Stage 1 和 Stage 2，Stage 3 和 Stage 4 的能力在 SBOM 生态里根本不存在。这是开源合规审计『制度供给 < 制度需求』的第一个精确量化——也是 Black Duck 退出→信通院行政标准真空的技术实证注脚。