DeepSeek Manipulates Information Pursuant to Chinese Law
It has been reported that the DeepSeek chatbot alters or suppresses responses to topics deemed politically sensitive by the CCP in 85% of cases, directly aligning outputs with Beijing’s censorship directives. This is not an accident—it is a calculated effort to expand the PRC’s control over global information. Moreover, it does this without any disclosure to users regarding how specific outputs have been altered pursuant to PRC law. Unlike American AI companies, which impose safeguards to limit genuinely harmful content, DeepSeek functions as a digital enforcer of the CCP, suppressing discussions on topics such as democracy, Taiwan, Hong Kong, and PRC human rights abuses. The model’s responses do not just echo Beijing’s messaging—they actively erase dissent, ensuring that only Party-approved narratives reach users.
Deepseek’s censorship operates on two levels: automated filtering erases responses before they even appear, while built-in biases systematically distort the AI’s overall behavior. The platform is designed to ensure the AI aligns with the CCP’s ideological and political objectives. PRC laws mandate that AI-generated content must reflect “core socialist values,” support “correct political direction,” and avoid material that could “incite subversion of state power.
Beijing also actively shapes how AI systems interpret, generate, and distribute information. Chinese regulations require firms to ensure algorithm “controllability” to give the PRC government direct influence over AI decisionmaking and allow authorities to modify AI behavior as needed. DeepSeek’s structure makes it inherently vulnerable to state manipulation, and without transparency into the extent of control exercised, its outputs must be assumed to serve Beijing’s strategic interests. Unlike AI models in open societies, DeepSeek exists in an ecosystem where compliance with state ideology is a prerequisite for survival. The result is an AI chatbot that cannot be trusted to provide an unbiased or unfiltered perspective, making it fundamentally compromised from its inception. To be clear, congressional efforts to date have focused on the data and security vulnerabilities associated with this app. However, it is important for the American people to also be aware of other challenges posed by the emergence of DeepSeek, such as its operator’s mandated compliance with PRC laws regarding covert information manipulation.
The danger is clear: millions of Americans are now using an AI system designed to serve the CCP. Beijing is not just censoring the internet at home. It is embedding its Great Firewall into platforms Americans use every day.
- DeepSeek’s Potential Unauthorized Distillation of U.S. AI Models
In the leadup to the release of its R1 model, there were allegations that DeepSeek engaged in a practice called “model distillation,” which involves the systematic extraction and replication of the reasoning capabilities of existing AI models to expedite their own development at reduced costs. The Select Committee—following meetings with a number of U.S. industry leaders—has determined that it is highly likely that DeepSeek used model distillation techniques to create an imitation AI model, copying leading U.S. AI models’ capabilities and violating U.S. companies’ terms of service.
Specifically, DeepSeek personnel infiltrated U.S. AI models and fraudulently evaded protective measures under aliases and purchased dozens of accounts using a sophisticated network of international banking channels. This allowed DeepSeek personnel to mask their identities, conceal their transactions, and avoid detection.
U.S. industry leaders told the Select Committee that they had “high confidence” that this has occurred. In one case, OpenAI wrote to the Select Committee that:
Through our review, we found that DeepSeek employees circumvented guardrails in OpenAI’s models to extract reasoning outputs, which can be used in a technique known as ‘distillation’ to accelerate the development of advanced model reasoning capabilities at a lower cost. Observations of DeepSeek’s R1 model also indicate instances of reasoning structures and phrase patterns that align with the behavior of OpenAI’s models. Additionally, we found that DeepSeek employees used OpenAI models to grade model responses and filter and transform training data, which are key steps in the AI development process. DeepSeek likely also used leading open-source AI models to create high-quality synthetic data.
Indeed, under Section 2—Usage Requirements—of OpenAI’s Terms of Use, the company expressly prohibits the use of its service to develop or improve competing modules. It states:
(f) You may not (i) use output from the Services to develop models that compete with OpenAI; (ii) use automated or programmatic methods to extract data from the Services; or (iii) use the Services to discover information about the models underlying the Services
OpenAI told the Select Committee that the first company to deploy a model replicating OpenAI’s o-series reasoning models was not a U.S. lab—it was DeepSeek
- DeepSeek’s Use of Export Controlled Nvidia Chips to Power its Model
DeepSeek’s AI model appears to be powered by advanced chips provided by American semiconductor giant Nvidia and reportedly utilizes tens of thousands of chips that are currently restricted from export to the PRC. Analytics firm SemiAnalysis has estimated that DeepSeek has at least 60,000 Nvidia chips, with orders for thousands more Nvidia H20 chips. The company estimates that DeepSeek is likely to have current access to the following Nvidia chips
• A100: 10,000
• H20: 30,000
• H800: 10,000
• H100: 10,000
Nvidia designed and manufactured many of these chips to create the most sophisticated possible chip while skirting U.S. export controls. This has allowed these chips to be exported to China as the U.S. government develops stricter restrictions. Since March 2024, it is estimated that Nvidia has produced over 1 million chips for the Chinese market
Setting aside the troubling practice of American companies deliberately and knowingly supplying the most advanced chips permissible under the U.S. export control regime to a foreign adversary, there is growing evidence of a coordinated effort by DeepSeek and other Chinese companies to violate U.S. law by illicitly importing banned chips into the PRC
In 2022, the U.S. Department of Commerce’s Bureau of Industry and Security announced sweeping export controls on advanced computing chips, looking to curb China’s access to the cutting-edge semiconductors crucial for artificial intelligence and military applications. Anticipating these new restrictions, Nvidia’s Chief Executive Officer, Jensen Huang, directed the company’s Head of Engineering for Graphics Processing to design a chip that would allow Nvidia to avoid the export control ban. Within a month of the export control announcement, Nvidia had successfully developed a modified chip that, according to analysts, was nearly as powerful as its top-tier processors at the time—undermining the intent of U.S. policy.
Another avenue in which DeepSeek and other Chinese firms gain access to highly sensitive chips is through intermediary countries that do not face the same export control restrictions as the PRC. These intermediary nations can serve as transshipment points, where companies can illegally purchase restricted chips before rerouting them to their final destinations.
Following DeepSeek’s release, Singaporean authorities charged three individuals—including one Chinese national—in connection with the illegal export of advanced Nvidia chips to DeepSeek in China in violation of U.S. export controls. Reports indicate that Singaporean law enforcement raided 22 locations and arrested at least nine individuals involved in the illicit network. These arrests occurred immediately after Chairman Moolenaar and Ranking Member Krishnamoorthi sent a bipartisan letter in January highlighting, among other matters, the threat of chip smuggling via Singapore
The Wall Street Journal further reported that “Chinese buyers are circumventing U.S. export controls to order Nvidia’s latest artificial-intelligence chips.” This aligns with growing concerns that Chinese companies, often with state backing, are systematically working to evade U.S. restrictions and continue acquiring advanced semiconductor technology. The U.S. Department of Commerce is currently investigating whether DeepSeek illegally imported exportcontrolled Nvidia chips from Singapore.
The Report concludes tomorrow
Illustration: Pixabay