DeepSeek breaks through in China and alerts the US: countermeasures
- Gabriele Iuvinale
- 24 ore fa
- Tempo di lettura: 8 min
The DeepSeek Chinese AI model has become popular in China, finding application in automobiles, mobile devices and health care. Western companies are also adopting its capabilities, but there is no shortage of security concerns
On January 20th, the Chinese artificial intelligence startup DeepSeek stunned the entire world by introducing its latest open source AI model able to challenge OpenAI's o1 model with its advanced capabilities: the DeepSeek-R1. Its success was immediately global. The Chinese model, in fact, represents one of the best open-source AI offerings on the global market. The V3 large language model also achieved performance levels comparable to those of major Western competitors, but at a much lower cost, as declared by the company.
At home, its use has spread like wildfire. DeepSeek, in fact, is now integrated into a wide range of government products and services in China.

In the rest of the world, it is also catching on. Some Western companies have already started adopting the R1 model which is now supported by cloud providers including Amazon and Microsoft.
However, there is no shortage of concerns about its use.
US policymakers and technologists have expressed concern that DeepSeek, a company with ties to the Chinese Communist Party (CCP) and the potential to share user data with China Mobile, could put US companies and national security at risk. In a new one political proposal, OpenAI - the Californian AI research laboratory - describes the Chinese startup DeepSeek as “state-subsidized” and “state-controlled” and recommends that the US government consider banning DeepSeek models and other similar Beijing-backed operations.
Success at home
More than 20 Chinese automotive brands have announced plans to incorporate DeepSeek models. Much of the adoption appears to be focused on improving existing AI features, including voice control, high-precision mapping, and smartphone-like access to music, web search, and messaging services.
Geely Auto, China's second-biggest automaker by sales, has declared that its latest models are able to understand even vague instructions from drivers, thanks to DeepSeek. A promotional video shows a driver saying "I'm a little tired" to the car's AI system. The DeepSeek-powered bot then instructs the car to close the windows, tilt the car seat to a 125-degree angle, dim the lights, and set an alarm to sound in 35 minutes to allow the driver to take a nap.
The company producing smartphone Huawei has used the R1 model to update its Siri-like AI assistant, Xiaoyi. The chatbot allows users to choose whether to respond to their requests with "automatic deep thinking" or "quick and short responses". The other large Chinese manufacturers such as Oppo, Honor, Vivo, Xiaomi have taken similar steps by incorporating features supported by R1, such as text and video creation and web search.
China's largest home appliance company, Midea, has launched a series of air conditioners powered by DeepSeek. The product is a “sympathetic friend” that can “accurately capture your thoughts,” claims the video of the company's launch.
Nearly 100 hospitals in China have announced they will adopt DeepSeek into their operations, according to a reported from Observer News. Applications include support for diagnostic and therapeutic processes, analysis of medical images, quality control of medical records and research into the use of new drugs.
DeepSeek has also been adopted by several Chinese government institutions. According to what reported from financial publication Caixin, the southern tech hub of Shenzhen was one of the first to use it in internal government systems. Longgang district has declared that during the Spring Festival, the entire staff of the Government Services and Data Management Bureau took the initiative to forego leave to complete the localized distribution of 671 billion parameters of the DeepSeek-R1 model and provide services for the region's more than 20,000 public officials.
DeepSeek in support of the Chinese government
Chinese institutions are adopting DeepSeek not only to improve their services, but also to signal their support for government goals, he said. declared at the Rest of World Chen Li, an analyst at the Beijing-based think tank Anbound.
“China has long identified AI as a strategic goal for the country's future development and wants to be a global leader in this sector... so pushing institutions and government offices to adopt AI is a very natural step towards achieving a greater goal,” he said.
This confirms how China is traveling fast towards global technological pre-eminence, positioning itself as a leading country in artificial intelligence (AI). In particular, Beijing has gained a competitive advantage in various aspects of the AI industry, showing strength in terms of academic papers, patents, cross-border investments and global financing.
Some Western companies have already started adopting DeepSeek's models, which are now supported by cloud providers including Amazon and Microsoft. As reported The Economist, Gloo, a messaging platform for churches founded by former Intel CEO Pat Gelsinger, used the model as the basis for its chatbot. Latenode, an automation platform, began offering R1 shortly after the model's launch, noting that its support for non-English languages was attractive to content marketers who wanted to translate material. Meta instructed the researchers to remove R1 and apply the Llama AI model.
The European banking sector is also interested in the Chinese DeepSeek model. According to the news site Information, Natwest and hsbc, two British lenders, are both experimenting with building their services on R1, as is Spain's bbva.
Western concerns
Politicians and technologists have expressed concern that DeepSeek has ties to the CCP, as well as the potential to share user data with China Mobile, which could put U.S. companies and national security at risk.
China Mobile, a state-owned telecommunications company and one of China's major mobile phone providers, has a code that is part of DeepSeek's login process. “The code within the DeepSeek chatbot contains heavily obfuscated scripts that, when deciphered, reveal a connection to computing infrastructure owned by China Mobile,” they write Lei. Nikakhtar died, Megan L. Marrone and Duane C. Pozza of the Wiley law firm. The first discoveries of data transmission to China Mobile's servers were discoveries from Feroot, a Canadian cybersecurity company.
Security concerns have prompted policymakers and stakeholders to widely discuss how to address DeepSeek R1. The reactions of Congress and the Administration reflect current policy debates about AI, national security, and economic competitiveness, the outcomes of which could impact many sectors of the economy and the U.S. innovation landscape.
OpenAI's censorship
In a new one political proposal, OpenAI describes DeepSeek as "state-funded" and "state-controlled", asking the US government to consider banning its use.
The proposal, presented for the initiative "AI Action Plan” of the Trump administration, claims that DeepSeek models, including the model R1 "reasoning", are insecure because they must comply with user data requests under Chinese law. Ban the use of models "made in the PRC" in all countries considered "Tier 1" according to the export rules of the Biden administration would avoid privacy and security risks,” OpenAI argues, including the “risk of intellectual property theft.” It is unclear whether OpenAI's references to “models” refer to DeepSeek's API, the lab's open models, or both, writes TechCrunch. “DeepSeek's open models contain no mechanisms that would allow the Chinese government to steal user data; companies like Microsoft, Perplexity, and Amazon host it on their infrastructures.”
OpenAI has already accused DeepSeek of "distilling" knowledge from OpenAI's models against its terms of service.
US censorship
Critics argue that DeepSeek users could unknowingly expose sensitive and proprietary information, from financial data to contracts and documents, raising privacy and national security concerns. US officials have expressed fears that if this data falls into the wrong hands, it could become a valuable asset to the CCP, given its history of exploiting such information.
Policymakers have proposed a number of responses to that Chinese model, many of which are noteworthy because they could have a broader impact on AI policy.
Government restrictions
The US government has taken steps to restrict DeepSeek. Several US federal agencies, including NASA, the Pentagon and the Navy, have banned its use on government-issued devices. This action mirrors previous actions against other Chinese apps, such as TikTok, and reflects ongoing concerns about data security and the potential for espionage.
On March 3, 2025, Representatives Darin LaHood (R-IL) and Josh Gottheimer (D-NJ) sent a letter to US governors to call for a ban on DeepSeek on government-issued devices. Several states have taken action in this direction: Virginia, Texas and New York have banned the use of DeepSeek on government systems.
Request one broader federal regulation
The US House Select Committee on the CCP has loudly expressed its concerns about DeepSeek and called for federal regulatory intervention. Committee chair John Moolenaar and ranking member Raja Krishnamoorthi sent a letter strongly addressed President Trump's National Security Advisor Michael Waltz, asking the administration to consider the possible national security benefits of introducing export controls on Nvidia's H20 and similar chips, arguing the importance of preventing PRC artificial intelligence systems from gaining market share in the United States. The Select Committee also recommended “updating the Federal Acquisition Regulations (FAR) to prohibit the federal government from purchasing PRC model-based AI systems such as DeepSeek, except for appropriate intelligence and research purposes.” In recent years, procurement regulations have become a powerful tool in trying to shape technology choices and supply chains. “Procurement limits, in the form of new contract clauses or certifications, can impact businesses across the economy and must be carefully monitored and evaluated,” they specify Here. Nazak Nikakhtar, Megan L. Marrone and Duane J. Pozza.
A legislative action by Congress
In addition to procurement, the US House Foreign Affairs Committee has raised concerns about violations of US export control laws, in light of the possibility that China gained access to Nvidia's restricted H100 chips through third countries. China's progress in DeepSeek has also raised broad concerns in Congress about Beijing's technological progress and the speed with which it is able to innovate.
These are the proposals:
R. 1121 - No DeepSeek on Government Devices Act. Representatives Gottheimer and LaHood introduced the bipartisan No DeepSeek on Government Devices Act, which would ban the use of DeepSeek by federal employees on government-issued devices.
R. 1122 - Technology Transfer Control Law of China. This bill, introduced by Mark Green (R-TN), directs the President to "regulate the export, re-export, or transfer" to China of any "technology of national interest or intellectual property" that falls within the jurisdiction of the United States or that has been exported by a U.S. individual.
S. 321: Decoupling Artificial Intelligence Capabilities from Chinese Law. Senator Josh Hawley (R-MO) introduced the Decoupling America's Artificial Intelligence Capabilities from China Act on January 29, 2025, to protect US AI development from China.
Trump has emphasized the importance of AI to economic and national security
One of President Trump's first moves on technology policy was the signing of theExecutive Order 14179 “Removing Barriers to American Leadership in Artificial Intelligence". The measure announces that the "policy of the United States is to support and strengthen the global dominance of AI in America to promote human well-being, economic competitiveness, and national security".
What to expect from the future
“AI innovators and users should watch the rapidly evolving regulatory landscape as a priority as policymakers weigh responses to DeepSeek and geopolitical economic competition,” they warn legal experts Nazak Nikakhtar, Megan L. Marrone and Duane C. Pozza.
Some export control regulations introduced by the Executive following the DeepSeek disclosure have been criticized as potential overreach by the US government. “There is always the risk of overcorrection, as it is difficult to develop regulatory programs that can precisely target one country, company or product without affecting others,” add the three experts. “AI users, including businesses, should heed government warnings and consider adjusting their AI governance to account for risks associated with specific models.”
The article was published in Italian here.
コメント