In the rapidly evolving landscape of artificial intelligence, a new rivalry has emerged that’s reshaping the entire industry. OpenAI, once the poster child for open-source AI research, has taken a controversial stance by proposing that the U.S. government consider banning models from Chinese AI lab DeepSeek. This move has ignited debate about the true meaning of “open” in artificial intelligence and highlighted the growing tensions between Western and Chinese AI development.
Let’s dive into this fascinating conflict and explore what it means for the future of AI development.
What’s Happening Between OpenAI and DeepSeek?
OpenAI recently submitted a policy proposal to the U.S. government that describes DeepSeek as a “state subsidized and state controlled operation.” The proposal recommends potentially banning models from DeepSeek and similar organizations supported by the Chinese government.
This recommendation comes at a time when DeepSeek has been gaining significant attention for its R1 model, which offers impressive capabilities while being more openly available than many Western counterparts.
The timing is particularly interesting given OpenAI’s recent pricing adjustment for their GPT-3.5 Mini model and renewed commitments to more open-source initiatives – moves that some industry observers interpret as direct responses to DeepSeek’s growing influence.
OpenAI’s Arguments Against DeepSeek
According to OpenAI’s proposal, they’re concerned about ensuring people have what they call “freedom of elligence” – the ability to access and benefit from advanced AI systems without interference from “autocratic powers” or excessive regulation.
OpenAI points out that more than 400,000 people worldwide use ChatGPT for various creative and productive purposes, positioning American AI development as competing against “a Chinese Communist Party determined to become the global Leader by 2030.”
The proposal specifically calls out DeepSeek’s R1 model as noteworthy not because of its capabilities (which OpenAI claims are “at best on par with several US models”) but because DeepSeek is:
- State subsidized
- State controlled
- Freely available
OpenAI argues that the cost to DeepSeek users is “their privacy and security,” suggesting that Chinese law requires DeepSeek to comply with government demands for user data.
The Privacy Paradox
One of the most glaring contradictions in OpenAI’s argument relates to privacy and security. The proposal suggests that using DeepSeek models compromises user privacy – but this claim ignores a fundamental difference in how these models can be used.
When you use OpenAI’s services, your data must pass through their infrastructure. Unless you have a special partnership with Azure, you can’t run OpenAI models locally in a way that fully protects your privacy.
In contrast, many AI enthusiasts run DeepSeek’s open models on their own hardware, completely eliminating the need to share data with any external entity – Chinese or otherwise.
As Theo put it, “When I’m running a model on my machines and my infrastructure, my privacy and security are not being compromised – unlike when I use OpenAI, where my data has to run on their systems.”
It’s like OpenAI warning you not to eat at a restaurant because they might spy on your conversation, while ignoring the fact that you can actually take their recipe and cook at home. Who’s listening to you in your own kitchen?
DeepSeek’s Innovations and Open Approach
What makes DeepSeek particularly remarkable isn’t just their models’ performance but their commitment to open research and innovation. In the past year alone, DeepSeek published 12 research papers detailing their training methods and technological breakthroughs.
These innovations are especially impressive considering the constraints Chinese AI companies face. Due to export restrictions, Chinese researchers don’t have access to the same high-bandwidth GPU connection technologies (like NVIDIA’s NVLink) that Western companies enjoy.
DeepSeek has developed novel approaches to minimize data transfer between GPUs, creating groundbreaking techniques that benefit AI researchers worldwide – regardless of geopolitical boundaries.
Their open-source contributions include:
- Custom file systems optimized for AI training
- Novel context compression techniques
- Efficient token generation methods
This level of transparency stands in stark contrast to the increasingly closed nature of most Western AI labs, including OpenAI itself.
The Irony of “Open” AI
Perhaps the most striking aspect of this controversy is the irony of a company named “OpenAI” advocating for restrictions on genuinely open AI research. While OpenAI has largely abandoned its original open-source mission (now reflected only in its name), DeepSeek has emerged as possibly “the most open AI company ever made.”
Industry observers note that Meta has been unusually quiet about their Llama models recently, with rumors suggesting they’re rebuilding much of their technology internally after seeing DeepSeek’s success. This speaks volumes about the quality and impact of DeepSeek’s open innovations.
Meanwhile, companies like Mistral AI, which position themselves as open alternatives, still maintain significant restrictions:
- Their models often can’t be run outside their infrastructure without substantial licensing fees
- They don’t open-source “the interesting stuff”
- There’s no way to host their models as efficiently as they do internally
I actually also had a “podcast” with Google’s NotebookLM agents regarding this very conversation and this was one of the main topics of the conversation, you can watch it here if you are curious
It’s like claiming your restaurant has an “open kitchen” because customers can peek through a tiny window, while your competitor is literally handing out all their recipes and cooking techniques for free.
The Reality Behind DeepSeek
Contrary to characterizations as a government operation, DeepSeek appears to have originated from private sector innovation. Many team members reportedly have backgrounds in cryptocurrency and were initially “just looking for things to do with their GPUs” when GPU mining became less profitable.
As one analysis points out: “This isn’t some government plant… No one expected DeepSeek to be as good as it is or to have found the success that they found – certainly not China, certainly not us, possibly not even them.”
If you wish to truly explore some of DeepSeek’s amazing contributions to the AI space, check out this great video on it (not affiliated as always):
The proposal acknowledges this reality in a single line, noting: “There isn’t any clear link between the Chinese government and DeepSeek” – a statement that seems to undermine the core premise of their entire argument.
The Real Concerns
While national security and privacy considerations in AI development are legitimate concerns, the specific proposal to ban DeepSeek’s models (rather than just placing restrictions on their APIs or cloud services) strikes many as overreaching.
The suggestion that individuals should be fined for simply downloading these models seems particularly extreme, akin to “Apple’s weird legal cases where they start suing people for importing motherboards to replace broken ones on phones because they had an Apple logo on it.”
A more nuanced approach might involve:
- Regulations on API access and cloud services
- Data protection requirements for cross-border AI services
- Promoting domestic open-source alternatives
- International cooperation on AI safety standards
The Road Ahead
As AI capabilities continue to advance globally, finding the right balance between openness, innovation, and security will be crucial. The tension between OpenAI and DeepSeek represents more than just corporate rivalry – it highlights fundamental questions about how AI development should proceed in an increasingly multi-polar world.
For AI enthusiasts and researchers, the silver lining may be that competition drives innovation. OpenAI’s GPT-3.5 Mini has been praised as “an incredible model” that many professionals rely on for solving complex problems, while GPT-4.5 represents “a great step in the right direction.”
Similarly, DeepSeek’s contributions to open AI research benefit the entire field, regardless of where you stand on the geopolitical spectrum.
Conclusion
The controversy surrounding OpenAI’s proposal to restrict DeepSeek models reveals the complex intersection of technology, commerce, and geopolitics in today’s AI landscape. While concerns about privacy, security, and international competition are valid, many in the AI community worry that overly restrictive approaches could hinder genuine innovation.
As Theo accurately noted: “Deep seek’s work should not be talked down on, it certainly should not be banned from the US, and it’s sad to see a company like OpenAI doing what they’re doing here.”
The true spirit of “open” AI may lie not in a company’s name but in its willingness to share knowledge, techniques, and innovations with the global research community. As the field continues to evolve, perhaps we’ll see a renewed emphasis on collaboration alongside the inevitable competition.
After all, as the saying goes, a rising tide lifts all boats – even if some of those boats are flying different flags.