⚠️ China Open Source is not origin open source, it’s ‘State-Chartered Codebase’, ‘Intranet Shared Source’, ‘Cyber-Estate/Bonsai Source’. If you read the news/article, please use these to replace the words ‘Open Source’.

China Open Source Daily — 2026-07-22

🏛️ Policy & Regulation

1. China Open Source Development Deep Report (2025) Released — AI Enters “Ecosystem Decisive” Era

The OpenAtom Foundation published the China Open Source Development Deep Report (2025), a comprehensive study based on data from GitHub, Gitee, and AtomGit. The report’s core thesis: global AI competition is shifting from “model比拼” to “ecosystem decisive”.

Key data points:

  • China’s open source contributions: 77.98 million contributions, up 54.92% year-over-year, accounting for 12.38% of global share (up from 10.10% in 2023), ranking third globally behind the US and India.
  • China’s active open source projects: 4.25 million, up 40.58% YoY.
  • Global AI open source projects: 790,200 active projects, up 70.59% YoY; active AI developers up 62.85%.
  • China’s AI open source projects: 95,200, accounting for 12.06% of global total, ranking third.
  • Chinese universities dominate the global Top 30 AI open source developer rankings: 15 out of 30 slots, with Tsinghua, Zhejiang University, Shanghai Jiao Tong, and Peking University in the top five globally.

The report makes a critical institutional observation: OSPOs are evolving from “compliance support departments” to organization-level infrastructure connecting enterprise R&D with external innovation ecosystems. In the AI era, OSPOs must now manage models, datasets, evaluations, inference frameworks, and application assets — not just software components and licenses.

The report also notes that open source infrastructure is shifting from “code repositories” to “intelligent platforms” — AtomGit is evolving into a “code + model + compute” integrated platform, representing a transition from “hosting projects” to “organizing innovation.”

Source: OpenAtom Foundation Journalism

2. AIP Open Source Project Launched Based on National AI Agent Interconnection Standard (GB/Z 185—2026)

On July 7, the AIP (Agent Interconnection Protocol) open source reference implementation was formally released on AtomGit, based on the newly published national standard GB/Z 185—2026 Artificial Intelligence — Agent Interconnection. The standard, comprising 7 guidance documents, addresses the lack of unified specifications for agent identity, capability description, discovery, interaction, and tool invocation across different vendors and platforms.

The AIP v2.1.0 reference implementation has already been downloaded over 6,000 times. The China Electronics Standardization Institute (CESI) and Beijing University of Posts and Telecommunications have jointly established the AIP open source community on OpenAtom’s open source operations platform, covering standards, technical sharing, SIG interest groups, community activities, and code repositories.

Institutional significance: This is a rare example of a national technical standard directly generating an open source reference implementation — a “standards co-development, open source, adaptation testing” three-in-one landing model that bridges the gap between regulatory specification and industrial practice.

Source: OpenAtom Foundation Journalism

⚖️ Legal & Licensing

3. OpenAtom Security Committee Hosts AI Security Forum — Supply Chain Governance in Focus

During the 2026 OpenAtom Open Source Ecosystem Conference, the OpenAtom Foundation’s Open Source Security Committee held a forum on “AI-era Security Ecosystem Research and Technology Analysis.” The forum addressed three major governance challenges:

  • AI security attack/defense frontiers: model poisoning, adversarial attacks, and AI vulnerability exploitation (with particular focus on the Mythos AI vulnerability framework)
  • Open source supply chain security: given AI systems’ deep dependence on open source models, training frameworks, third-party libraries, and plugins, the forum emphasized building trusted open source code bases with full dependency tracking and poisoning detection
  • Infrastructure and OS security: extending security from application-layer protection to operating systems, runtime environments, toolchains, and compute infrastructure

Security Committee Chair Wu Jingzheng noted that “AI security governance must shift from after-the-fact response to built-in, lifecycle security assurance.”

Source: OpenAtom Foundation Journalism

🏗️ Institutional Change

4. OpenAtom Foundation Adds New Partners (June 2026)

The OpenAtom Foundation announced new institutional partners for June 2026:

Platinum Donors (foundation level):

  • Shenzhen Kaihong Digital Industries — Deepin/OpenHarmony ecosystem platform company, a major OpenHarmony code contributor with 73 compatibility-certified products, 4 PMC seats, 35 Committers, and 330 dedicated code contributors
  • Loongson Technology — China’s leading domestic CPU designer
  • TalentedSoft Information Systems — IoT solutions provider

Open Source Contributor:

  • Shenzhen Micro-Nano Integrated Circuit & Systems Application Technology Research Institute

Community-level donors:

  • OpenHarmony: Chongqing Chuanyi Automation (A-class)
  • openKylin: China Great Wall Technology (Platinum), Glanf Intelligent Tech (Silver)
  • OpenAtom Electric IoT: Wasion Information Technology (Gold)

Institutional significance: The addition of Loongson (LoongArch CPU architecture) and Shenzhen Kaihong (Deepin/OpenHarmony ecosystem) as platinum donors signals OpenAtom’s strategy of building a vertically integrated Chinese open source stack — from Loongson’s indigenous CPU architecture up through the operating system layer.

Source: OpenAtom Foundation Journalism

5. OpenAtom Foundation Adds Three New Incubation Projects (June 2026)

Three new projects formally entered OpenAtom’s incubation pipeline:

  • Anolis OS (Dragon Lizard) — The OpenAnolis community’s operating system, positioning as a “next-generation Agentic OS open source root community” with an “AI×OS×Cloud” roadmap. The community has 20,000+ developers, 343,000+ users, and enterprise deployments at XPeng, Zeekr, OPPO, and China Unicom.
  • xLLM — Proposed by JD.com (Beijing Wodong Tianjun Information Technology), targeting resource efficiency and SLO guarantees in large-scale LLM deployment.
  • LoongArch Binary Translator (LAT) — Proposed by Loongson Technology, enabling binary translation for the indigenous LoongArch instruction set architecture.

Source: OpenAtom Foundation Journalism

6. OpenTenBase Mid-Year Event — 500,000+ Nodes Deployed, Institutional Milestones

The OpenTenBase mid-year event was held in Beijing, attended by over 100 guests from academia, industry, and the open source community. Key milestones:

  • 50,000+ nodes of OpenTenBase kernel deployed in production
  • OpenTenBase 5.0 and TXSQL 8.0.30 dual-version major release, with HTAP architecture upgrade and LLM integration
  • Academician Wang Huaimin (CAS) sent a congratulatory message emphasizing the importance of open source in database R&D
  • Professor Zhou Aoying (ECNU, CCF Database Committee Chair) noted that OpenTenBase is “China’s first PostgreSQL-based distributed open source database project entrusted to a neutral foundation” — establishing a governance benchmark for domestic foundational software
  • Two new project donation agreements were signed, expanding the ecosystem

Source: OpenAtom Foundation Journalism

🔍 Cultural & Geopolitical Signals

7. Xinhua: AI Is Redefining Open Source — From Code Sharing to “Full Domain Ecosystem Reconstruction”

An article republished by the OpenAtom Foundation from Xinhua News Agency argues that AI has fundamentally broken the traditional boundaries of open source:

“In the past, open source was about source code sharing, developer collaboration, and software version iteration. Today, open source must simultaneously address models, data, compute, agents, terminals, business models, and governance rules.”

The article reports that China’s open source root communities (openKylin, openEuler, OpenHarmony, LoongArch, OpenTenBase) are transitioning from individual efforts to a coordinated “phalanx effect” — building a full-stack autonomous foundational software system covering terminal, edge, cloud, and network.

On commercialization, the article identifies three emerging models: MaaS (API-based billing), private deployment + vertical fine-tuning, and “free base + paid premium” hybrid models. The AaaS (Agent as a Service) model is expected to grow the Chinese AI agent market from ¥18.2 billion (2025) to ¥3.3 trillion (2028).

Source: OpenAtom Foundation Journalism (originally from Xinhua)

8. Xiaomi’s openvela: Open Source as AI Hardware Base

At the 2026 OpenAtom Open Source Ecosystem Conference, Xiaomi Open Source Committee Chair Zhang Duo announced that openvela — Xiaomi’s lightweight IoT operating system — is positioning as the “intelligent base” for AI hardware. Key institutional observations:

  • 1.8 billion devices shipped with Xiaomi Vela technology (watches, earphones, speakers, smart glasses, etc.)
  • Apache 2.0 licensed, open sourced on GitHub and Gitee in late 2024
  • Smallest footprint: runs on 32KB RAM BLE modules; supports up to 256MB RAM smart devices
  • First MCU OS to pass POSIX PSE52 certification
  • AI Native development system: knowledge base, programming Agent, 15 AI Skills covering the full lifecycle, 80+ debugging tools
  • End-side AI inference framework for real-time, privacy-preserving local AI

Institutional significance: Xiaomi’s decision to open source a platform validated on 1.8 billion devices represents a shift from proprietary IoT fragmentation to open ecosystem play — a strategy that mirrors Google’s Android playbook but applied to the AI hardware layer.

Source: OpenAtom Foundation Journalism

🔍 WeChat Monitor (Institutional Signals)

9. OpenAtom Electric IoT: “RunDianHong” Powers Grid Digitalization

China Resources Power (华润电力) released “RunDianHong” (润电鸿), the first enterprise distribution of the OpenAtom Electric IoT (电鸿) operating system, targeting renewable energy station intelligent inspection. Deployed across 900+ standardized smart terminals in Shandong province, the system achieved:

  • Equipment inspection coverage: from 70% → 95%+
  • AI defect recognition accuracy: 92%
  • Per-person daily inspection range: from 5km → 20km
  • Fault localization time: from 30 minutes → 1-3 minutes
  • Annual operating cost savings per station: ¥650,000+

Source: OpenAtom Foundation Journalism

10. openEuler 24.03 LTS SP4 Released — Deepening AI and Full-Scene Innovation

openEuler 24.03 LTS SP4 was released, based on the 6.6 kernel, with enhancements for server, cloud, and AI scenarios — including kernel optimization, NPU compute partitioning, inference service fast recovery, sandbox, intelligent diagnostics and tuning, compiler, and confidential virtual machines. The release was contributed by 2,006 developers from 39 member organizations.

Source: OpenAtom Foundation Journalism

11. DORA-rs Community Deeply Engages at OpenAtom Ecosystem Conference — Embodied Intelligence Robotics

The DORA-rs open source community, a high-performance Rust-based robotics middleware project incubated by OpenAtom, hosted an “Embodied Intelligent Robotics Modern Software Architecture Hands-on Workshop” at the 2026 OpenAtom Ecosystem Conference. The workshop demonstrated full integration of Rust-native robotics middleware with the Octos intelligent agent framework, contributing to China’s embodied intelligence open source ecosystem.

Source: OpenAtom Foundation Journalism

🔍 Commentary

The OpenAtom Foundation’s journalism page — which aggregates 12 substantive articles published between July 2 and July 16 — offers a uniquely rich window into the institutional structure of Chinese open source. Several patterns merit attention:

Pattern 1: The “Phalanx Effect” is Real. The Xinhua article accurately describes the transition from individual projects to coordinated “root community” strategy. The simultaneous releases from openEuler (SP4), OpenTenBase (5.0), and the new incubation of Anolis OS, xLLM, and LoongArch LAT — all under OpenAtom’s neutral governance — represent a deliberate institutional architecture. The Chinese open source stack is being assembled as a vertically integrated, horizontally coordinated ecosystem, not a collection of isolated projects.

Pattern 2: Standards as Institutional Levers. The AIP project (national AI agent standard → open source reference implementation) demonstrates a distinct Chinese approach to standard-setting: standards are co-developed with open source implementations from the start. This contrasts with the Western model where standards bodies (ISO, IETF, W3C) and open source projects operate in parallel tracks. The “standards + open source + certification” three-in-one model is a genuine institutional innovation in how China bridges regulatory governance with technical practice.

Pattern 3: OSPO Evolution is Under-Reported. The Deep Report’s observation that OSPOs are evolving from compliance departments to “organizational infrastructure” for AI innovation ecosystems is significant. If Chinese OSPOs are indeed expanding their remit to manage models, datasets, and AI assets alongside traditional software components, this represents a convergence with global OSPO best practices — but with a distinct Chinese emphasis on “open source as industrial policy instrument.”

Pattern 4: The Great Divergence in AI Open Source. The report’s data on Chinese universities (15 of top 30 global AI open source developer slots) versus the US emphasis on “large project pools” reveals a structural divergence: China is building a deep developer talent pool at the university level, while the US maintains advantages in original project creation and global community influence. The question is whether developer quantity can translate into project quality and community leadership — or whether the institutional environment (academic incentives, industry collaboration models, IP frameworks) will constrain the conversion.

Pattern 5: Open Hardware Strategy Takes Shape. Xiaomi’s openvela open-sourcing (1.8 billion validated devices → Apache 2.0) and Loongson’s platinum donor status at OpenAtom suggest an emerging open hardware + open software strategy for AI. Unlike the Android model (Google controls the platform, OEMs customize), China’s approach appears to be: anchor with indigenous CPU architectures (LoongArch), provide open-source operating systems at all layers (OpenHarmony for IoT, openKylin for desktop, openEuler for cloud, openvela for AI hardware), and govern through a neutral foundation.

🇺🇸🌏 Geopolitical Escalation

1. US Treasury Secretary Threatens Sanctions on Chinese Open-Source AI Models over IP Theft

In a major escalation of the US-China AI technology competition, Treasury Secretary Scott Bessent said on July 21 that the Trump administration will “carefully examine” Chinese open-source AI models for signs of intellectual property theft, and could impose sanctions. Speaking at a press conference, Bessent specifically cited concerns that Chinese AI companies may have distilled proprietary American models to build their open-weight releases — a direct reference to the Kimi K3, DeepSeek-V4, and other frontier-class open-source models that have dominated headlines in recent weeks. The administration is reportedly exploring whether sanctions under existing executive orders on technology theft can be applied to model weights distributed as open source.

This marks the first time the US government has explicitly threatened to sanction open-source AI model weights, potentially setting a precedent that would fundamentally alter the legal landscape of open-source AI distribution globally. CNBC, Bloomberg, TechCrunch, and Business Insider all covered the announcement.

Source: TechCrunch, Bloomberg, CNBC, Business Insider

2. MIT Technology Review: China’s AI Models Have Trump’s AI World at War with Itself

MIT Technology Review published a deep analysis on July 20 examining how the wave of Chinese open-weight AI models — particularly Kimi K3 — has fractured the Trump administration’s AI strategy. The report describes a White House divided into factions: those who advocate for restricting Chinese AI models through export controls and sanctions, versus those who argue that open-source competition from China demonstrates the futility of proprietary US AI business models. The article notes that “every time a new smart, free model from China like Kimi gets released, US companies see less reason to fork out money to access” proprietary American systems. This internal division mirrors the broader structural tension in US AI policy: whether to compete through openness or restrict through control.

Source: MIT Technology Review

3. White House Export Controls on US AI Labs Could Paradoxically Boost Open Source

A related analysis by CryptoBriefing (June 29) argues that the Trump administration’s export controls on frontier AI models from Anthropic and OpenAI may inadvertently strengthen global open-source AI. By restricting the export of proprietary US models, the US government is pushing international developers toward Chinese open-weight alternatives — the very thing it seeks to contain. The article notes that the administration’s approach creates a “self-defeating cycle” where restrictions on US AI exports accelerate the adoption of Chinese open-source models abroad.

Source: CryptoBriefing

🇺🇳🏛️ International Governance

4. China Explicitly Champions Open-Source AI at UN’s First Global Dialogue on AI Governance

While the world’s attention was on WAIC 2026 in Shanghai, an equally significant institutional development occurred two weeks earlier in Geneva. At the UN’s first Global Dialogue on AI Governance (July 6-7), China’s Minister of Industry and Information Technology Li Lecheng delivered a keynote speech explicitly championing open-source AI as a global public good. He cited DeepSeek and Qwen by name, declaring: “Open source AI is a shared asset for all humanity. Chinese open source models such as DeepSeek and Qwen have significantly lowered the barriers and costs of AI adoption.”

This is the first time a Chinese government official has explicitly named specific Chinese open-source AI projects in a formal UN setting. The speech made three proposals: (1) AI for all — bridging the Global South AI divide; (2) open source cooperation and innovation; (3) jointly building global AI governance. China also submitted a formal written response to the UN’s AI governance questionnaire, explicitly listing “open-source software, open data and open AI models” as a priority area for the dialogue.

Notably, China’s submission also called for opposing “drawing ideological lines, forming exclusive blocs, or fragmenting global industrial and supply chains under the guise of AI governance” — a direct rebuttal to the US export control regime.

The Geopolitechs analysis notes an interesting tension: the same Chinese government that is championing open-source AI at the UN has also been reported to be studying export controls on its own frontier models. The article concludes that the UN speech represents the authoritative policy position, and the export control discussions are largely irrelevant.

Source: Geopolitechs, China.org.cn, UN Global Dialogue on AI Governance

🔍 Commentary

The Sanctions Paradox: How IP Theft Accusations May Redefine Open Source

The convergence of three developments this week creates a defining moment for open-source AI:

  1. Bessent’s sanctions threat against Chinese open-source AI models represents the first time a major government has directly threatened the legal status of open-source model weights. From an institutional economics perspective, this is an attempt to redefine property rights over AI model weights — extending intellectual property enforcement into a domain that the open-source community has treated as a commons. If successful, this would create a new category of “sanctionable open source” — an oxymoron that would fundamentally challenge the premise of open-source distribution.

  2. China’s UN positioning as the champion of open-source AI creates a remarkable geopolitical role reversal. The country that the US accuses of IP theft is now the primary advocate at the UN for open-source AI as a “shared asset for all humanity.” This is a classic institutional entrepreneurship move: by defining open-source AI as a global public good, China positions its own model releases as contributions to the commons, while framing US restrictions as attempts to hoard technology.

  3. The MIT Technology Review analysis reveals the internal contradiction in US AI policy: the administration simultaneously wants to (a) restrict Chinese AI models through sanctions, (b) export controls on US AI companies, and (c) maintain US dominance in AI. These three goals are mutually inconsistent. Export controls on US AI push international users toward Chinese alternatives; sanctions on Chinese models create a precedent that could be used against any open-source distribution; and neither approach addresses the fundamental competitive dynamic: Chinese companies are winning global developer mindshare by giving away frontier-class models.

The paradox is captured perfectly by the timing: as Treasury Secretary Bessent threatens sanctions on Chinese open-source models, the UN’s first Global Dialogue on AI Governance has already institutionalized China’s position as the leading state advocate for open-source AI. The battle over open source is no longer just about code — it is about who gets to define the governance framework for the most important technology of the decade.

📊 Trends

1. The Sanctions Precedent for Open-Source Weights The US Treasury’s threat to sanction Chinese AI models over IP theft would, if realized, create a new legal category: “sanctioned open-source software.” This would have profound implications for platforms like Hugging Face, GitHub, and any model registry that hosts weights from sanctioned entities. The open-source community must now grapple with a reality where model weights are treated as strategic assets subject to trade controls.

2. UN as a Governance Arena for Open Source China’s proactive engagement at the UN Global Dialogue on AI Governance signals a strategic shift: rather than merely participating in technical open-source communities, China is now using intergovernmental forums to establish its preferred norms for open-source AI governance. The explicit naming of DeepSeek and Qwen in a ministerial speech at the UN is unprecedented and signals that open-source AI is now a formal component of China’s foreign policy.

3. The Fragmentation of US AI Policy The MIT Technology Review analysis reveals a US AI policy apparatus in internal conflict. The “containment faction” (sanctions, export controls) and the “competition faction” (open-source competition, deregulation) are pulling in opposite directions. This institutional fragmentation creates uncertainty for both US and Chinese open-source communities — the rules of the game are unsettled, and the cost of regulatory risk is rising.

4. The Role Reversal in Open-Source Advocacy The geopolitical dynamics of open-source advocacy have inverted. China, once seen as a net consumer of Western open-source software, is now the leading state-level advocate for open-source AI at the UN. The US, home of the open-source movement, is now leading efforts to restrict open-source distribution of AI models. This institutional role reversal has profound implications for the global governance of open source.