Do Chinese open-source AI models actually contain backdoor risks?

There is currently no publicly available independent verification evidence supporting claims that mainstream Chinese open-source models contain backdoors. Open-source code can be audited by anyone, which actually reduces the possibility of hidden backdoors. This section of the committee report is more based on hypothetical scenarios and inference than verified technical fact.

Why do the committee's proposed countermeasures generate controversy?

The main points of controversy are: restricting open-source AI tools may hinder American technological innovation itself; government subsidization of American open-source AI has questionable efficacy; overly aggressive restrictions could trigger trade friction and put American tech companies using Chinese tools for secondary development in a difficult position.

What does the rise of DeepSeek and Qwen mean for American AI companies?

From a competitive standpoint, this represents real pressure—high-quality, low-cost alternatives compress pricing space for OpenAI, Anthropic and others in mid-to-lower market segments. But from a technological perspective, competition tends to drive innovation. The rise of Chinese open-source AI is also forcing American companies to accelerate open-source strategies—Meta's Llama series is a case where competitive pressure has driven continuous strengthening.

US Commission Warns China Open-Source AI Threatens American AI Leadership

In March 2025, the U.S. Congressional Select Committee on the Chinese Communist Party released a strongly-worded warning report, directly alleging that China's open-source AI ecosystem is eroding American global leadership in artificial intelligence through a "dual-track strategy." The report focuses on Chinese open-source model series including DeepSeek and Qwen, contending that these models—through extremely low cost barriers, competitive technical benchmarks, and permissive licensing—are rapidly penetrating the global AI developer community and effectively building new AI infrastructure anchored on Chinese technical standards.

The committee's core concerns span three dimensions: technology diffusion risks (Chinese models may contain backdoors or covert data collection mechanisms), accelerated talent drain (China's cost-effective tools reduce global developer dependency on American AI toolchains), and standards-setting authority (whoever's open-source ecosystem achieves wider adoption gains greater influence over AI industry norms and interface standards). The report recommends Congress restrict federal procurement of Chinese AI tools and support American open-source AI projects through subsidies.

The report's timing is notable, arriving as global discussion of DeepSeek R1 had not yet subsided and Qwen series models were continuing to climb open-source model leaderboards. While parts of the report's reasoning face scrutiny from AI researchers—since open source implies code auditability, which actually reduces backdoor risk—the geopolitical technology competition it describes is real, and the struggle for AI infrastructure dominance between the U.S. and China is entering a new phase.

US Committee Warns: China's Open-Source AI "Dual-Track Threat" to American Leadership

Report Background: A Geopolitical Manifesto from Capitol Hill

In March 2025, the U.S. Congressional Select Committee on the Chinese Communist Party (SCCCP) released a comprehensive report titled "Open Source Artificial Intelligence and China's Technology Strategy." At over 80 pages, this represents the committee's most detailed and systematic policy document on AI to date, drawing on evidence and analysis from government intelligence agencies, academic researchers, and technology industry experts. The report makes an unambiguous assertion: through its open-source AI strategy, China is systematically dismantling the technological advantages the United States has built in artificial intelligence over many years.

The bipartisan committee, established to address "the systematic challenge posed by the Chinese Communist Party to American security, economy, and democracy," has previously issued similar warnings about quantum computing, semiconductors, and 5G. The current AI report represents the latest extension of this ongoing technology competition policy action series.

What is the "Dual-Track Strategy"? The Committee's Core Allegations

The report's central concept is the so-called "dual-track strategy": one track involves "commercial and academic penetration"—maximizing global technological influence by releasing high-performance AI models in open-source form; the second track involves "civil-military fusion"—these same models and underlying technologies can be leveraged by China's defense industry and intelligence agencies to support military AI capabilities.

The report identifies the following Chinese AI products as primary concerns:

DeepSeek Series: DeepSeek sent shockwaves through the global AI community with its remarkable cost-effectiveness. Its R1 model matches OpenAI's o1 in reasoning capabilities but reportedly costs 1/50th as much to run inference on. The committee's concern extends beyond performance to DeepSeek's alleged hidden ties to Chinese government subsidy programs.

Qwen Series: Alibaba Cloud's Qwen series has become a fixture at the top of global open-source model leaderboards, with multiple versions ranking among the most downloaded on HuggingFace. The report notes that Qwen model user agreements contain data repatriation clauses that may provide Alibaba Cloud (and by extension Chinese authorities) with global user behavioral data.

Baichuan, InternLM, and Others: The report enumerates several additional Chinese open-source model projects, arguing they collectively constitute a coordinated "technology ecosystem strategy."

Three Threat Dimensions: Detailed Analysis

#### Dimension One: Technical Security Risks—Backdoors and Data Sovereignty

The report cites security researcher analyses to advance the following risk scenarios:

**Covert Data Collection**: Model integration layers (inference APIs) may transmit user inputs back to Chinese servers via hidden network requests
**Backdoor Vulnerabilities**: Model weights may contain embedded trigger words that, when activated, generate specifically biased outputs or execute particular actions
**Intellectual Property Traps**: Some Chinese models may have distilled knowledge from American companies' proprietary models without authorization, creating legal exposure for users

Notably, a significant portion of these allegations are hypothetical and currently lack publicly verifiable technical evidence. Multiple AI security researchers have raised objections, arguing that the auditability of open-source code is precisely its security advantage, and that the "backdoor" argument contains logical flaws.

#### Dimension Two: Talent and Ecosystem Competition

The report's second line of attack is more substantiated. It uses detailed data to demonstrate the penetration speed of Chinese AI tools in the global developer community:

Downloads of models from Chinese institutions on HuggingFace grew 340% over the previous 18 months
International contributors to AI projects with Chinese-language documentation on GitHub grew by over 200%
In emerging markets including Southeast Asia, South Asia, and Latin America, adoption rates of Chinese AI tools (especially Qwen and DeepSeek) have surpassed American equivalents

The substantive significance of this penetration is that once developers build workflows around Chinese AI tools, switching costs become extremely high. This resembles a "technology platform lock-in" effect—similar to American concerns about Chinese telecommunications equipment, Chinese AI tools could become digital infrastructure that becomes difficult to replace.

#### Dimension Three: The Covert Battle for Standards-Setting Authority

This is the most strategically profound and abstract section of the report. It notes that standards-setting authority has historically been closely tied to the dominant technology ecosystem—TCP/IP originated from American research institutions, HTTP/HTML was led by CERN, and smartphone operating system standards were defined by Apple and Google. AI-era "standards" encompass: model interface specifications, prompt engineering paradigms, evaluation benchmark systems, and safety boundary definitions.

If a billion developers worldwide are using Chinese open-source AI models, they will naturally adapt to these models' output styles, API design logic, and content moderation boundaries. This invisibly provides Chinese technology companies with influence over the next generation of AI standard-setting.

Responses from All Sides: Consensus and Controversy

Voices Supporting the Committee's Position:

Some national security experts argue that even if technical risks are overstated, geopolitical risks are real—countries dependent on Chinese AI infrastructure will be in weaker negotiating positions when Sino-American relations are tense
DARPA has reportedly advised in internal memos that defense-related contractors limit use of Chinese AI models

Voices Questioning the Report's Arguments:

Hugging Face co-founder Clement Delangue publicly stated that "open source transparency is precisely its security guarantee; closed source should be the regulatory priority"
Researchers at AI safety organization METR note that no independently verified evidence currently exists demonstrating that mainstream Chinese open-source models contain malicious backdoors
Some American technology companies (particularly those using Qwen as a foundation for secondary development) privately oppose overly aggressive procurement restrictions

Policy Recommendations and Market Implications

The report advances six policy recommendations to Congress:

1. Legislatively prohibit federal government agencies from deploying Chinese-developed AI models in core systems

2. Establish AI provenance certification standards (an AI-sector extension of "Buy American" legislation)

3. Provide dedicated federal subsidies of no less than $5 billion to American open-source AI projects

4. Require AI products sold in the U.S. to disclose supply chain provenance

5. Coordinate with allies to establish a "Trusted AI Supply Chain" alliance

6. Increase intelligence monitoring of Chinese AI research and development activities

Deep Analysis: Fear, Reality, and Strategic Anxiety

This report reflects the deep anxiety within America's policy elite about the potential loss of technological hegemony. In a sense, the rise of Chinese open-source AI challenges an assumption Americans once took for granted: that frontier technology would only flow from the United States to the world, not in reverse.

DeepSeek R1 shattered this presumption—a model trained under conditions of constrained chip resources matched or exceeded American top models on multiple benchmarks. This made Washington's policy circles realize that export controls may not be able to maintain the technology gap, and in some areas China may have already established independent technological pathways.

This anxiety is real, but converting it into effective policy faces fundamental challenges: how to find an executable balance between supporting open innovation and guarding against geopolitical risks? The report does not provide a convincing answer to this question—but its very existence signals that the AI-specific phase of U.S.-China technological competition has arrived in earnest.

Furthermore, this development signifies a broader transformation in the global technology landscape. As governments and enterprises worldwide continue to increase their investments in AI technologies, the accompanying legal and regulatory frameworks are also evolving rapidly. Regulators face the ongoing challenge of balancing innovation promotion with user protection.

From an industry perspective, these changes are driving increased cross-sector collaboration and integration. The convergence of traditional industries with AI technologies is creating new business models and value chains. Simultaneously, talent development and technology standardization have become critical factors for industry advancement.

Looking ahead, we can anticipate continued rapid development in this field, bringing forth more innovative applications and solutions. Companies must maintain acute market awareness and adjust their strategic directions promptly to adapt to emerging technological trends. User experience enhancement and data security assurance will become core competitive elements.

The implications extend beyond immediate technical considerations to encompass broader socioeconomic impacts. Organizations must prepare for significant shifts in operational paradigms while ensuring sustainable and responsible implementation of these technologies.