Three weeks ago, millions of users around the world eagerly downloaded the DeepSeek application, an AI chatbot touted as a more cost-effective and powerful alternative to OpenAI’s ChatGPT.

With its advanced capabilities, lower cost, and open-source accessibility, DeepSeek quickly became the most downloaded free app on Apple’s App Store, triggering excitement and anxiety in equal measure.

However, soon after its widespread adoption, several countries – including Italy, South Korea, and Taiwan – began imposing bans on the Chinese AI platform, citing national security concerns and data privacy risks.

The latest to join the growing list is the US, where the states of Texas, New York, and Virginia have prohibited government employees from downloading and using DeepSeek on state-owned devices and networks.

The move follows similar restrictions in Europe, Australia, and parts of Asia, as Western governments question the security implications of allowing a Chinese AI model to collect and process user data.

Despite the bans, experts argue that DeepSeek marks a significant challenge to the US’ longstanding dominance in the AI sector. Some even suggest that Washington and its allies are reacting out of fear rather than genuine security threats.

“The combination of low development costs, faster responses, accuracy, affordability, and open-source accessibility makes DeepSeek an attractive alternative to more established generative AI tools such as ChatGPT,” said Harin Sellahewa, dean of the Faculty of Computing, Law, and Psychology at the University of Buckingham.

“The instant popularity of this unexpected new entrant shook the AI industry and even caused stock markets to tumble,” he told Anadolu.

How is DeepSeek different?

DeepSeek claims to have built its chatbot with a fraction of the budget and resources typically required to train similar models. While leading AI firms use over 16,000 high-performance chips to develop their models, DeepSeek reportedly used just 2,000 older-generation chips and operated on a budget of less than $6 million.

Explaining the platform’s underlying technology, Sellahewa said: “DeepSeek, like OpenAI’s ChatGPT, is a generative AI tool capable of creating text, images, programming code, and solving mathematical problems. Both models, known as Multimodal Large Language Models (MLLMs), are trained using vast amounts of data from diverse fields.”

However, what sets DeepSeek apart is its use of the Mixture of Experts (MoE) architecture, which enables the AI model “to consult many experts from various disciplines and domains” within its framework to generate a response.

“In terms of accuracy, DeepSeek’s responses are generally on par with competitors, although it has shown to be better at some tasks, but not all,” he continued.

“For example, DeepSeek has shown to be better at solving math problems than providing detailed summaries of articles with broad contextual information.”

Additionally, he added, DeepSeek has positioned itself as an open-source AI model, meaning developers and researchers can access and modify its algorithms, fostering innovation and expanding its applications beyond what proprietary models like ChatGPT allow.

Deepak Padmanabhan, a senior lecturer at the School of Electronics, Electrical Engineering, and Computer Science at Queen’s University Belfast, also believes that DeepSeek is not radically different from other chatbots in terms of functionality.

“Its key distinction lies in its higher energy efficiency and lower operational costs,” he said.

Why did DeepSeek cause a stir?

DeepSeek’s rapid adoption and performance against competitors such as OpenAI and Google sent shockwaves through the tech industry.

Following its launch, major US tech stocks saw sharp declines, with chipmaker Nvidia alone reportedly losing nearly $600 billion in market value.

For Dana McKay, associate dean of Interaction, Technology & Information at the Royal Melbourne Institute of Technology (RMIT), DeepSeek’s impact can be attributed to two key factors.

“Firstly, it is being billed as open source, because it allows certain configurations, though the underlying code and dataset have not been released,” she said.

“Second, it is owned by a Chinese company, and that comes with certain laws behind it – user data must be stored in China, and the Chinese government can then access it for a variety of reasons, including economic reasons.”

Padmanabhan believes that DeepSeek’s emergence represents more than just a technological leap – it is a geopolitical challenge.

“DeepSeek is pretty much the first big chatbot from outside the American Big Tech sector … from outside the confines of the US,” he said.

It is even more significant given the US restrictions on exporting advanced Nvidia chips to China, which was expected to cripple China’s AI industry, he said.

Which countries are banning DeepSeek and why?

Several countries have moved swiftly to ban or restrict DeepSeek, particularly for government employees.

Italy was the first country in Europe to remove the chatbot from app stores, citing concerns over how user data was collected, stored, and used. The Italian data protection authority has formally asked DeepSeek to disclose what personal data it collects, from which sources, and where it is stored.

South Korea, Australia, and Taiwan have also barred government officials from using DeepSeek due to security risks.

Last week, the Netherlands became the latest to ban DeepSeek for civil servants, citing potential espionage threats. Dutch officials expressed concerns that the Chinese government could use the AI platform for surveillance or cyber-espionage.

Padmanabhan believes the bans could have various reasons, but one significant driver is geopolitics.

“With Europe, Australia and UK largely being culturally and economically aligned with the US, they see DeepSeek as a significant threat to Western dominance,” he said.

Such bans, he added, could have unintended consequences, and lead to more curiosity and public interest in DeepSeek.

Are DeepSeek security concerns legitimate?

China has criticized the bans as an “overstretching” of the concept of national security, with the Foreign Ministry reiterating that Beijing “attaches great importance to data privacy and security and protects it in accordance with the law.”

McKay believes that there are valid security risks associated with the use of DeepSeek.

“Given the large amount of data it collects – what people are looking for, information about their device and where it is located, their keystroke patterns – there is a plausible national security risk,” she said.

“This level of information could lead the Chinese government to know an awful lot about the movements of individuals, or what they are looking for. Keystroke data, for example, can be used to uniquely identify individuals even where they are accessing DeepSeek using a different account.”

In Sellahewa’s view as well, there are legitimate concerns over DeepSeek’s data collection and privacy policy.

The privacy policy allows it to collect user-provided information such as usernames, email addresses, and phone numbers, he said.

“Moreover, the automatically collected data includes keystroke patterns or rhythms, which can be used as a biometric to identify individuals,” he said.

DeepSeek also has legal obligations to comply with Chinese government requests, meaning that user data stored on its servers could be accessed by authorities, he said.

“Given that information collected by DeepSeek is stored in servers located in the People’s Republic of China, personal data of users outside of China might not be protected by data protection regulations they might normally expect,” he added.

Do Western AI models do the same?

At the same time, the experts pointed out that Western AI models such as ChatGPT and Google’s Gemini also collect vast amounts of user data, raising questions about whether bans on DeepSeek are justified solely on security grounds.

“OpenAI ... isn’t explicit about collecting, e.g. keystroke level data, but does collect a lot of data and feed it back into their models,” said McKay.

Padmanabhan agreed, asserting that all generative AI technologies, not just DeepSeek, collect data at “a very fine-grained level.”

All generative AI technologies use extensive data surveillance, so there are security and privacy risks for every user, he added.

“It is not very obvious as to why DeepSeek collecting data is qualitatively any different from OpenAI or Google collecting data,” he said.

The academic also emphasized that most uses of the technology are informational or entertainment-oriented, so “it is not conceivable that these could directly link with national security.”

“These concerns are valid only if the users think that companies in one country are more trustworthy than others, which brings us back to the issue of geopolitics, and the West-East conflicts within the contemporary global political order,” he said.