This Thursday, OpenAI submitted a policy proposal to the US government that has drawn a lot of attention and generated a lot of clickbaity headlines. Many journalists focused on the fact that the policy proposal directly attacks DeepSeek, an OpenAI competitor.
But the real points people should be taking away from the proposal are more nuanced. This isn’t just a case of OpenAI attacking a competitor or trying to erect favorable regulatory moats around its business (although that could certainly be one of the underlying motivations of the proposal). Instead, the proposal speaks to the crossroads the US, and the world, faces with AI.
It originated as a response to the White House’s call for public comment on an AI Action Plan that the government is putting together as the result of an executive order issued in late January aimed at “removing barriers to American leadership in artificial intelligence”, which called on the Office of Science and Technology Policy to submit an action plan to the President within 180 days. OpenAI’s proposal had several key points:
AI progress is accelerating. OpenAI points to the fact that “the cost to use AI falls by 10x every 12 months”, that “the amount of calendar time it takes to improve an AI model keeps decreasing”, and “AI models are catching up with human intelligence at an increasing rate.” It also stated that it believes we are “approaching artificial general intelligence”.
AI should be shaped by democratic principles. OpenAI argues that we must ensure “freedom of intelligence” by guarding against “autocratic powers that would take people’s freedoms away” and the “layers of laws and bureaucracy” that would prevent AI progress in democratic countries.
China is “determined to overtake us by 2030”. OpenAI says that in order to prevent China from overtaking the US, the US needs to do a number of key things: “export democratic AI”, increase partnership between the private sector and the federal government on AI, remove regulatory barriers around copyright for AI training data, increase investment in AI infrastructure, export “democratic AI” and promote the global adoption of American AI systems, and ban Chinese AI companies like DeepSeek.
Throughout the proposal, there’s an emphasis on the urgent threat posed by China. OpenAI paints a dark picture of what it calls Chinese “authoritarian AI”, which the CCP can use as a tool “to amass power and control their citizens”. It also makes the argument that China’s authoritarian regime gives it a number of “strategic advantages”, including the ability to quickly marshal resources to subsidize its AI industry, immunity to western IP laws which frees up its ability to use data to train its models, and “regulatory arbitrage” being created by the 781 state-level AI bills introduced in 2025 alone that could dampen American AI innovation.
In particular, OpenAI singles out DeepSeek as representative of “authoritarian AI” and its risks. The proposal says that “DeepSeek is simultaneously state-subsidized, state-controlled, and freely available, the cost to its users is their privacy and security” and that “[DeepSeek’s] models also more willingly generate how-to’s for illicit and harmful activities such as identity fraud and intellectual property theft, a reflection of how the CCP views violations of American IP rights as a feature, not a flaw.”
OpenAI also repeatedly mentioned Huawei in the same breadth as DeepSeek, clearly alluding to the precedent set by the US government’s numerous bans on the Chinese telecom company due to national security concerns. These bans began in 2017 and escalated in scope through 2022, spanning both the Trump and Biden administrations. OpenAI’s intent in mentioning Huawei so frequently was clearly to suggest similar actions should be taken on DeepSeek.
OpenAI also suggested a number of so-called “freedom-focused policy proposals” to countermand China’s AI advantages and “advance democratic AI”:
First, it proposed a framework for “voluntary partnership between the federal government and the private sector” on AI.
Second, it proposed export controls targeted at restricting “the flow of AI technologies to the PRC”, with a three-tiered system to govern AI diffusion.
Thirdly, it argued that fair use copyright laws should apply to AI companies using copyrighted data to train their data, arguing that “unpredictable availability of inputs hinders AI innovation, particularly for smaller, newer entrants with limited budgets.”
Finally, it said that the US should increase investment in AI infrastructure, arguing that “hundreds of billions of dollar sin global funds are waiting to be invested in AI infrastructure” and that if the US doesn’t channel those resources into its own projects, “the funds will flow to projects backed and shaped by the CCP”.
Some may point to the apparent contradiction in OpenAI simultaneously calling for export restrictions and bans on foreign competitors like DeepSeek (i.e. greater regulation and government involvement) while also arguing for the removal of regulatory barriers like copyright laws as a case of wanting to have their cake and eating it too.
However, the policy proposal makes a compelling case that US national security interests are aligned with OpenAI’s recommendations. Irrespective of OpenAI’s ulterior motives for the message it’s sending, it is choosing the right time to send it and the right government to send it to. The Overton window around protectionism and free trade has been smashed wide open by the new administration in recent months, and by positioning its policy proposal as a battle plan for the widening AI front in the escalating trade wars surrounding the US, it wouldn’t be surprising to see many of its recommendations echoed in the final version of the AI Action Plan that the Trump Administration ultimately adopts come July.
