In the unfolding contest over who will shape the intelligence of machines, OpenAI has asked the U.S. government to ban Chinese AI models — most notably DeepSeek — framing the request as a matter of national security and intellectual sovereignty. The proposal arrives weeks after DeepSeek outpaced ChatGPT on app store rankings, achieving comparable performance at a fraction of the cost and headcount, a fact that has led thoughtful observers to ask whether the call for regulation is born of principle or of pressure. The episode surfaces a tension as old as competition itself: the line between pro
OpenAI seeks ban on Chinese AI models while keeping own systems closed-source
Advocating for bans while keeping its own systems closed
Mark
So OpenAI is asking the government to ban Chinese AI models. On its face, that sounds like a national security argument. What's the actual concern they're raising?
Mimi
They're saying that because DeepSeek operates under Chinese law, the government could force it to hand over user data. And then that data gets used to train better AI systems for the Chinese state. It's a data sovereignty issue, not just a capability issue.
Luke
Right, but here's what's fuzzy: OpenAI's proposal doesn't clearly distinguish between DeepSeek's API—which could theoretically be monitored by Beijing—and DeepSeek's open-source models, which are hosted on Microsoft and Amazon servers. Those are different security profiles. The proposal lumps them together.
Mark
So the security risk might not apply equally to all of DeepSeek's offerings?
Mimi
Exactly. If you download an open-source model and run it locally, there's no data flowing back to China. The risk OpenAI describes—compelled data sharing with the government—doesn't exist in that scenario.
Luke
And that's where the competitive angle becomes hard to ignore. DeepSeek just beat ChatGPT on the app store. It's doing more with fewer people and lower costs. OpenAI has 2,000 employees; DeepSeek has 160.
Mark
So is this about security or about market share?
Mimi
It could be both. The security concerns might be real. But the timing and the breadth of the proposal—banning all PRC-affiliated models—suggest that competitive pressure is also in play.
Luke
And there's the irony: OpenAI keeps its own models closed-source, which limits transparency and independent security auditing. Meanwhile, it's asking the government to ban open-source models from China. That's a hard position to defend on principle.
Mark
What does DeepSeek say about all this?
Mimi
The founder says they're independent, lean, focused on breakthroughs not money. He met with Xi Jinping recently, which obviously fuels suspicion. But there's no hard evidence that DeepSeek is a state tool.
Luke
And that's the thing—the evidence is murky. We know DeepSeek is effective and cheap. We know the founder has ties to Chinese leadership. We don't know whether those ties mean the government controls the company or just that the founder is prominent enough to meet with Xi.
Le Pouls
- DeepSeek stunned the AI world by surpassing ChatGPT on app store charts with roughly 160 employees and a fraction of OpenAI's resources, exposing a vulnerability OpenAI cannot easily ignore.
- OpenAI's 15-page White House proposal names DeepSeek as a state-linked security threat, warning that Chinese law could compel the company to funnel user data to government authorities.
- Critics have seized on a glaring contradiction: OpenAI keeps its own models closed and proprietary while demanding bans on open-source rivals, raising questions about whether security or competitive survival is the real motive.
- The ambiguity in OpenAI's proposal — it never clarifies whether it targets DeepSeek's API, its open-source models, or both — leaves the actual security risk poorly defined and the regulatory ask dangerously broad.
- With its $500 billion Stargate Project and deepening Washington influence, OpenAI is no longer just a participant in AI governance debates — it is actively writing the rules that could determine who competes and who is shut out.
In the unfolding contest over who will shape the intelligence of machines, OpenAI has asked the U.S. government to ban Chinese AI models — most notably DeepSeek — framing the request as a matter of national security and intellectual sovereignty. The proposal arrives weeks after DeepSeek outpaced ChatGPT on app store rankings, achieving comparable performance at a fraction of the cost and headcount, a fact that has led thoughtful observers to ask whether the call for regulation is born of principle or of pressure. The episode surfaces a tension as old as competition itself: the line between protecting a commons and protecting a market share is rarely drawn by disinterested hands.
OpenAI has submitted a 15-page policy proposal to the White House calling for the U.S. government to ban AI models developed by entities affiliated with China, with DeepSeek named as the primary target. The company argues that Chinese law could compel DeepSeek to share user data with state authorities, and that freely available Chinese models could help Beijing develop more capable AI systems for government use. The proposal also takes aim at the 781 state-level AI bills introduced this year, which OpenAI argues could fragment innovation and weaken American competitiveness, while advocating for infrastructure investment, copyright reform, and expanded access to government data.
The timing has drawn immediate scrutiny. In January, DeepSeek surpassed ChatGPT on app store rankings, and its V3 model outperformed several leading U.S. systems — including GPT-4o — on third-party benchmarks, all while operating with roughly 160 employees compared to OpenAI's 2,000-plus workforce. The cost and efficiency gap has led observers to question whether OpenAI's regulatory push reflects genuine security concerns or a bid to neutralize a formidable rival through policy rather than competition.
The contradiction embedded in OpenAI's position has not gone unnoticed. The company maintains closed, proprietary models while calling for restrictions on open-source alternatives like DeepSeek's — a stance critics describe as seeking protection for itself while limiting access for others. Notably, companies including Microsoft, Amazon, and Perplexity host DeepSeek's open models on their own infrastructure, where the data-sharing mechanisms OpenAI warns about do not apply.
DeepSeek's actual relationship to the Chinese government remains murky. It originated as a spin-off from a quantitative hedge fund, and its founder has publicly characterized the lab as lean and independent. No direct state ties have been established, though a meeting between the founder and Xi Jinping shortly before OpenAI's proposal has sharpened the scrutiny. OpenAI has also accused DeepSeek of improperly extracting knowledge from its models — an allegation that marks a significant escalation in their dispute.
What the proposal leaves unresolved is perhaps its most important flaw: it never specifies whether the ban would apply to DeepSeek's API, its open-source models, or both. That ambiguity matters enormously, because the security risks of each are not equivalent. As the U.S. government weighs these proposals, the central question is whether OpenAI's security arguments are strong enough to stand on their own — or whether they are inseparable from the company's interest in eliminating a competitor it could not outpace on merit alone.
OpenAI has asked the U.S. government to ban artificial intelligence models developed by entities affiliated with China's government, specifically naming DeepSeek as a target. The request came in a 15-page policy proposal submitted to the White House as part of its call for public input on an AI Action Plan. OpenAI characterized DeepSeek as state-controlled and state-subsidized, arguing that its models pose privacy and security risks because Chinese law could compel the company to share user data with government authorities. The company also warned that DeepSeek's freely available models allow the Chinese government to access training data that could be used to develop more capable systems for state use.
The proposal extends beyond DeepSeek. OpenAI urged the U.S. government to consider banning AI models from any entity with People's Republic of China backing in countries classified as "Tier 1" under current export rules. The company framed this as necessary to prevent intellectual property theft and to maintain American competitiveness. OpenAI also criticized the 781 state-level AI bills introduced in the current year, arguing they could stifle innovation and weaken the U.S. position against China's AI ambitions. Beyond restrictions, the proposal advocates for increased infrastructure investment, copyright reform, and greater access to government datasets for AI development.
The timing and framing of OpenAI's push have drawn scrutiny. DeepSeek made headlines in January when it surpassed ChatGPT on app store rankings. The Chinese lab's V3 model outperformed several top U.S. AI systems—including Meta's Llama 3.1, OpenAI's GPT-4o, and Alibaba's Qwen 2.5—on third-party benchmarks, and it did so at significantly lower cost. DeepSeek operates with approximately 160 employees, compared to OpenAI's workforce of more than 2,000. This disparity has led some observers to question whether OpenAI's regulatory push stems from genuine security concerns or from competitive pressure.
The contradiction at the heart of OpenAI's position has not escaped notice. While the company keeps its own models proprietary and closed-source, it is actively calling for restrictions on open-source models like those from DeepSeek. Critics have pointed out this asymmetry: OpenAI seeks regulatory exemptions for itself while pushing for bans on rivals, and it advocates for transparency and fairness in AI governance while maintaining opacity about its own systems. DeepSeek's open-source approach, by contrast, allows broader access for developers and researchers. Companies including Microsoft, Perplexity, and Amazon host DeepSeek's open models on their infrastructure, and these hosted versions do not include mechanisms that would allow the Chinese government to extract user data.
DeepSeek's relationship to the Chinese government remains unclear. The company began as a spin-off from High-Flyer, a quantitative hedge fund. While no direct evidence links DeepSeek to the Chinese state, founder Liang Wenfeng met with Chinese leader Xi Jinping a few weeks before the regulatory proposal, a detail that has intensified scrutiny. Wenfeng has publicly distanced DeepSeek from China's major technology companies and venture capital firms, characterizing the lab as lean, mission-focused, and independent. He has stated that the company prioritizes breakthroughs over monetization and is not driven by commercial pressures. OpenAI has previously accused DeepSeek of improperly extracting knowledge from its models, a charge that represents an escalation in the companies' dispute.
What remains unresolved is the scope of OpenAI's concerns. It is unclear whether the company's objections target DeepSeek's API, its open-source models, or both. The proposal does not specify. This ambiguity matters because open-source models hosted on third-party infrastructure operate differently from proprietary systems, and the security risks may not be equivalent to those OpenAI describes.
The broader stakes are significant. OpenAI's $500 billion Stargate Project and its growing influence in Washington mean the company is not merely participating in AI governance debates—it is helping to shape the rules that will govern the industry. The conflict between OpenAI and DeepSeek raises fundamental questions about protectionism versus security, about whether open-source development strengthens or weakens innovation, and about who should have authority to set the terms of global AI competition. As the U.S. government considers these proposals, it will need to weigh OpenAI's security arguments against the possibility that regulatory action serves primarily to eliminate a formidable competitor.
Citations marquantes
America is competing with a CCP determined to become the global leader by 2030. DeepSeek's R1 model is noteworthy not because of its capabilities but as a gauge of the state of this competition.— OpenAI, in policy proposal to White House
We are in the early stage of a revolution. Companies should focus on breakthroughs instead of monetization.— Liang Wenfeng, DeepSeek founder
Envie de l'histoire complète ?
Lire l'original sur TechStartups.com ↗
Politics & Governance