China Open Source Daily — 2026-07-24

⚠️ 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 — 2026-07-24

🏛️ Policy & Regulation

1. MIIT Releases First Official Quantification of China’s Open Source Ecosystem — 1.35 Billion OpenHarmony Devices, 10 Billion Global AI Model Downloads

At the State Council Information Office press conference on July 20, MIIT spokesperson and Director of the Operation Monitoring and Coordination Bureau Tao Qing provided the first comprehensive official quantification of China’s open source ecosystem during the 2026 H1工业和信息化发展情况 (Industrial and Information Technology Development) briefing.

Key data points released by MIIT:

Tao Qing specifically addressed the question of “how to cultivate the open source ecosystem” in response to a journalist’s query, making four key points:

1. Basic software accelerating: OpenHarmony now covers smartphones, computers, cars, and home appliances — a full-stack coverage from IoT to desktop
2. Industrial software widely deployed: Coverage across key industries, reaching 86.3% digital R&D tool adoption
3. AI fully empowering: Software enterprises adopting code LLMs and intelligent programming tools at scale
4. Open source ecosystem accelerating: The national foundation has incubated 50+ projects; the national AI open source community now hosts 70,000+ models with 11M+ users

Open source as explicit policy priority: Tao Qing stated that MIIT will:

• Accelerate the introduction of an “AI + Software” action plan (人工智能+软件行动方案)
• Promote intelligent transformation of software development, smart upgrade of software products/services, and cultivation of AI agent software as a new business category
• Enhance open source infrastructure in project incubation, supply chain security governance, emerging scenario exploration, and overseas market landing
• Accelerate the development of the national AI open source community into a “source of innovation for emerging technologies and an important scenario for software-hardware product deployment”

Institutional significance: This is the first time China’s government has released comprehensive, quantified data on its open source ecosystem through a formal State Council press conference. The framing is significant: open source is presented as an integrated component of industrial policy — not merely a development methodology but a strategic infrastructure for software, AI, and hardware ecosystem development. The explicit mention of “overseas market landing” as a policy priority for open source infrastructure signals that Chinese open source is now a vehicle for global technology deployment, not just domestic innovation.

Sources: China News Service, AIBase (English summary), State Council Information Office

2. “AI + Software” Action Plan — A New Industrial Policy Instrument for Open Source

Beyond the data, the MIIT press conference revealed a significant new policy instrument: the “AI + Software” action plan (人工智能+软件行动方案). Unlike previous “Software + AI” framings which treated AI as a bolt-on feature, the new formulation positions AI as the primary driver of software industry transformation.

Three pillars of the action plan:

• Software development intelligent transformation: Code LLMs, AI programming tools, and automated testing/operations
• Software product and service intelligent upgrade: Smart features, natural language interaction, AI-generated content
• AI agent software as new business category: The “AaaS” (Agent as a Service) model — consistent with the ¥3.3 trillion AI agent market projection cited in the Xinhua/OpenAtom analysis from July 22

The action plan is explicitly linked to open source infrastructure: Tao Qing stated that the national AI open source community will be the “source of innovation” for the plan. This creates a direct institutional link between open source governance and industrial policy implementation — a distinctive Chinese approach that differs from the Western model where open source and industrial policy operate in separate domains.

Source: China News Service

⚖️ Legal & Licensing

3. White House Specifically Accuses Moonshot AI of Distilling Anthropic’s Fable 5 for Kimi K3 — Treasury Sanctions Threat Intensifies

On July 22-23, the White House escalated its confrontation with Chinese open-source AI by making specific allegations against Moonshot AI’s Kimi K3 model:

Michael Kratsios, Director of the White House Office of Science and Technology Policy (OSTP), publicly accused Moonshot AI of using covert distillation of Anthropic’s Fable 5 to train Kimi K3 — a 2.8-trillion-parameter open-weight model released on July 16. Kratsios also alleged that Moonshot accessed Nvidia’s Blackwell GPUs despite export controls.

US Treasury Secretary Scott Bessent doubled down on his sanctions threat (first reported July 21), stating that sanctions remain on the table and that the administration is “carefully examining” Chinese open-source AI models for IP theft.

Key details of the allegations:

• Kratsios stated on X that the US government “holds information confirming Moonshot AI distilled capabilities from Anthropic’s Fable AI during development of Kimi K3”
• The accusation draws a line between open-weight releases (which the administration says it supports) and covert extraction of proprietary model outputs (which it calls theft)
• Moonshot’s K3 is scheduled for full open-weight release on July 27
• Anthropic’s Fable 5 was re-released on July 1 after being taken offline due to US export controls, giving a narrow window for distillation claims
• Some AI researchers have questioned the claims, noting that Kimi K3’s architecture differs significantly from Fable 5

Institutional significance: The framing by Bessent is strategically important — he explicitly distinguishes between “legitimate open-source AI” and “stolen IP released as open source.” This is an attempt to create a legal distinction that would allow the US to sanction specific open-weight models without appearing to oppose open source in general. If this distinction becomes legally operationalized, it would create a new category of “tainted open source” — models that are open in form but sanctioned in substance. The burden of proof shifts to model creators to demonstrate that their training data and methods are clean, an extremely difficult standard for open-weight releases.

Sources: TechCrunch, CNBC, The Register, PCMag, ExplainX, ZeroHedge

🏗️ Institutional Change

4. Jensen Huang Defends Chinese Open-Source AI — A Pivotal Intervention from the Industry’s Most Influential CEO

In an exclusive interview with Axios on July 22, Nvidia CEO Jensen Huang made a forceful defense of Chinese open-source AI models, directly challenging the US administration’s escalating sanctions campaign:

“These Chinese models are excellent. Open-source models that are excellent should be used.”

Key arguments from Huang’s interview:

• Openness = security: Huang argued that open models are more secure, not less, because outside researchers can inspect them, expose weaknesses, and build defenses
• Monoculture risk: “If everything just becomes one single model, one single point of attack, one single source of failure, I think the world is much less safe” — a direct argument against the proprietary model approach
• Market reality: Chinese open-source models are already widely used by enterprises globally, and bans would disrupt existing deployments
• Business logic: Open-source AI expands the overall market for AI infrastructure, including Nvidia’s chips and data centers — a self-interested but structurally significant argument

Institutional significance: Huang’s intervention is remarkable for several reasons:

1. Nvidia’s strategic position: Nvidia is the primary beneficiary of the AI infrastructure buildout on both sides of the Pacific. Huang has more to lose from AI fragmentation than almost any other industry figure
2. The timing: His statement comes just as the White House and Treasury Department are escalating sanctions — a direct counter-signal from the industry’s most powerful voice
3. The framing: Huang doesn’t argue that Chinese models are not competitive with US models (which would be defensive). He argues they are excellent and should be embraced — an offensive reframing that positions openness as a strength, not a vulnerability
4. The global audience: The interview was published in Axios, Fortune, Yahoo Finance, and MoneyControl — reaching both US policymakers and global investors

This is the most significant industry voice to push back against the US sanctions campaign against Chinese open-source AI, and it signals that the “containment faction” and “engagement faction” in US AI policy now have a major corporate champion.

Sources: Axios, Fortune, Yahoo Finance, MoneyControl, TimesNow

5. Arcee (US Open Source AI Lab): “Chinese Models Are Not Inherently Dangerous”

On July 22, Arcee — a US-based open-source AI company — published a notable counterpoint to the Washington panic. CTO Lucas Atkins stated clearly:

“Chinese models are no more dangerous than any other open source software a company may use.”

Key arguments:

• Download-and-run security: Once a model is downloaded and run in an enterprise’s own environment, the original developer has no access to it — the model is just weights, not a remote service
• Open source != espionage: Open-weight models cannot “phone home” or exfiltrate data unless the user explicitly configures them to do so
• Inspection advantage: Open-source models can be audited, inspected, and modified by the user — unlike proprietary models where the internal workings are opaque
• Practical benefit: Arcee itself uses Chinese open-source models as base models for fine-tuning, finding them technically competitive with US alternatives

Institutional significance: Arcee’s statement is notable because it comes from a US company that competes in the same market as Chinese AI labs. Unlike Jensen Huang (who has commercial interests in chip sales), Arcee’s argument is purely technical and practical. The company is effectively saying: “The open-source model is working as designed — we inspect, we use, we build on top. The security concerns are not technical but political.”

This represents a growing divide between US policy makers (who view Chinese open-source AI as a national security threat) and US technology practitioners (who view it as a useful resource). The Arcee position, combined with Jensen Huang’s, suggests that the US open-source AI community is actively resisting the securitization of open-weight model distribution.

Sources: TechCrunch, BitcoinWorld, HotON.ai

🔍 WeChat Monitor

6. OpenAtom Foundation Journalism — MIIT Press Conference Coverage

The OpenAtom Foundation’s official website prominently features the MIIT press conference results on its homepage, framing the 1.35 billion OpenHarmony devices and 10 billion AI model downloads as validation of the foundation’s incubation model. The foundation’s journalism page has been updated with the July 20 press conference as the lead story, signaling that the foundation views this official quantification as a significant legitimization event.

Source: OpenAtom Foundation

🔍 Commentary

The Week of Official Quantification: Three Signals for Chinese Open Source

1. The MIIT Press Conference Represents a Governance Phase Transition

The July 20 press conference is the first time China’s government has released quantified open source data through a formal State Council briefing. This is not a technical report from a standards body — it is a ministerial-level policy statement that treats open source as a formal category of industrial output, measured alongside software revenue (¥6.2 trillion) and digitalization rates (86.3%).

The institutional significance cannot be overstated: open source has been officially recognized as a measurable component of China’s technology economy. The metrics — 1.35 billion devices, 10 billion downloads, 50+ incubated projects, 11 million platform users — create a baseline against which future policy will be evaluated. This is the same governance logic that underlies the OpenAtom Deep Report (covered July 22): what gets measured gets managed.

2. The “AI + Software” Plan Creates a New Institutional Link

The “AI + Software” action plan, with its explicit linkage to the national AI open source community, represents a Chinese innovation in industrial policy architecture. In the Western model, open source foundations and industrial policy operate in separate spheres — foundations govern code, governments set market conditions. In the Chinese model, the national AI open source community is explicitly designated as the “source of innovation” for a government industrial policy action plan.

This is neither top-down control (the community has 11 million users, not a government-appointed board) nor bottom-up autonomy (the policy explicitly directs funding and attention to the platform). It is a hybrid governance model that the Chinese system is uniquely equipped to implement: a government-funded, community-operated, industry-oriented open source platform that serves as a policy instrument without being a government agency.

3. The Convergence of the Sanctions Narrative and the Quantification Narrative

This week presents a remarkable juxtaposition: on the same days that the US Treasury threatens sanctions on Chinese open-source AI models (July 21-22) and the White House specifically accuses Moonshot AI of theft (July 22-23), the Chinese government publishes its first official quantification of the open source ecosystem it has built (July 20).

The MIIT’s 10 billion global download figure for Chinese AI open-source models is a direct counterpoint to the US narrative. The data says: these models are being used, everywhere, by everyone — they are not theft, they are adoption. The 29 nations that signed WAICO (covered July 23) are the diplomatic manifestation of the same argument: Chinese open-source AI is a global phenomenon, not a covert operation.

The institutional insight: China is now fighting the narrative war with data, not just rhetoric. The State Council press conference provided the data; WAICO provided the diplomatic vehicle; the Mulan License donation (covered July 23) provided the legal infrastructure. The three developments together form a coherent institutional response to the US sanctions campaign: quantify, legitimize, govern.

4. The Industry Counter-Narrative Takes Shape

Jensen Huang and Arcee represent a significant development: the US technology industry is beginning to push back against the sanctions narrative. Two arguments are emerging:

• Security through openness (Huang): Open models are more secure because they can be inspected. Banning them makes the ecosystem less safe.
• Technical normalcy (Arcee): Chinese open-source models are just software — no more dangerous than any other open-source dependency.

These arguments are structurally important because they align the interests of the US open-source community with the Chinese open-source ecosystem. If the US sanctions campaign succeeds, it will harm not only Chinese AI labs but also US companies that build on Chinese open-source models. The industry pushback suggests that the “bilateral enclosure of the open-source AI commons” (identified in the July 22 commentary) has a powerful constituency on the US side that will resist it.

📊 Trends

1. The Official Quantification of Chinese Open Source The MIIT press conference establishes a new governance baseline: Chinese open source is now measured, reported, and managed as a formal component of technology industrial policy. Expect annual or semi-annual updates to these metrics, creating a data series that will inform both domestic policy and international positioning.

2. The “AI + Software” Action Plan as a Policy Instrument This is the first time a major Chinese government action plan explicitly designates an open source community as its institutional foundation. If successful, this model could be replicated for other technology domains (AI + manufacturing, AI + healthcare, etc.), creating a new category of policy-aligned open source infrastructure.

3. The Industry-Policy Divide in US AI Governance The divergence between Jensen Huang/Arcee (industry) and the White House/Treasury (policy) on Chinese open-source AI represents a growing institutional fracture. The US has no equivalent of China’s centralized quantification or coordinated policy response — instead, it has competing factions with incompatible interests. This institutional fragmentation is a structural advantage for China’s coordinated approach.

4. The “Tainted Open Source” Legal Precedent If the US Treasury establishes a sanctions regime that distinguishes between “legitimate” and “tainted” open-source model weights, it will create a legal category with no precedent in open-source governance. The open-source community has never had to certify that its training data was not “stolen” — the concept is foreign to the legal framework of open-source licensing. A sanctions regime that requires such certification would fundamentally alter the governance of open-weight AI distribution.