Although the Chinese artificial intelligence startup DeepSeek has emerged as a formidable challenger to Western AI giants, causing ripples across the global tech landscape and raising concerns about national security and information control.

DeepSeek’s latest AI model, unveiled in late January 2025, has demonstrated capabilities that rival or even surpass those of industry leaders like OpenAI, Meta, and Google. What sets DeepSeek apart is its claim of achieving these results at a fraction of the cost and computational power typically required for such advanced AI systems.

DeepSeek’s Key Developments

1. Cost-Effective Innovation: DeepSeek’s R1 model reportedly cost just $6 million to train, approximately 95% less than its Western counterparts. This cost efficiency has sent shockwaves through the industry, causing significant drops in stock prices for companies like NVIDIA.
2. Performance Metrics: The new AI model has matched or exceeded the performance of OpenAI’s GPT-4 on several math and reasoning metrics, challenging the assumption of Western dominance in AI technology.
3. Open-Source Approach: Unlike some of its competitors, DeepSeek has published its methodologies and made its models freely available to researchers worldwide, fostering rapid advancements in the open-source AI community.
4. Technological Breakthroughs: DeepSeek’s efficiency gains are attributed to innovative approaches, including a “mixture of experts” architecture and the use of synthetic training data generated from existing AI models.

What data is DeepSeek collecting?

According to DeepSeek’s privacy policy, the company collects a wide range of personal data, including:

• Profile information: Username, email, phone number, password, and date of birth.
• User input: Everything you type or upload, including chat history, prompts, and audio input.
• Device and network data: IP address, device model, operating system, system language, and keystroke patterns.
• Usage data: Features you use, actions you take, and system performance logs.
• Cookies and trackers: Web beacons and other tracking technologies to monitor user behavior.
• Third-party data: Information from linked accounts and advertising partners that track your activity across websites, apps, and stores.

Where does DeepSeek store user data?

DeepSeek stores user data on servers located in China. This practice has raised alarms among European regulators and cybersecurity experts worldwide. The storage of data in China is particularly concerning due to the country’s National Intelligence Law, which compels companies to cooperate with government intelligence efforts without legal recourse to resist.

Is there any safe way to use DeepSeek?

Given the extensive data collection practices and the storage of data in China, experts advise caution when using DeepSeek. Some recommendations for safer usage include:

1. Limit personal information: Avoid sharing sensitive or personally identifiable information when using the service.
2. Use a VPN: Consider using a virtual private network to mask your IP address and location.
3. Separate accounts: If you must use DeepSeek, create a separate email account for registration to minimize data linkage.
4. Regular deletions: Frequently delete your chat history and any stored data within the app.
5. Stay informed: Keep up with the latest security updates and regulatory actions regarding DeepSeek.

However, it’s important to note that even with these precautions, the fundamental concerns about data storage and potential government access remain.

DeepSeek’s Comparative Analysis

DeepSeek has boldly positioned its new R1 model as a direct competitor to industry leaders, particularly OpenAI’s models. On their website, DeepSeek provides a comprehensive comparison that highlights the strengths of their latest offering:

1. Performance Parity: DeepSeek claims that their R1 model’s performance is on par with “OpenAI-o1,” likely referring to GPT-4. This assertion is supported by benchmark results showcased on their website, demonstrating comparable or superior performance across various tasks.
2. Open-Source Advantage: Unlike many of its competitors, DeepSeek has fully open-sourced the R1 model and its technical report. This move towards transparency allows researchers and developers to examine and build upon their work, potentially accelerating AI advancements.
3. Licensing Freedom: DeepSeek has taken a significant step by releasing R1 under the MIT license. This liberal licensing allows for free distillation and commercialization, potentially disrupting the AI market by enabling wider adoption and innovation.
4. Cost-Effectiveness: The company presents a pricing structure that appears to be more competitive than its rivals. For instance, they offer rates of $0.14 per million input tokens (for cache hits) and $2.19 per million output tokens, positioning themselves as a more affordable option in the market.
5. Technical Innovations: DeepSeek highlights the use of large-scale reinforcement learning in post-training, which they claim has led to significant performance improvements with minimal labeled data. This approach has reportedly allowed them to achieve parity with leading models in math, code, and reasoning tasks.
6. Distilled Models: In addition to their main R1 model, DeepSeek has open-sourced six smaller, distilled models. They assert that their 32B and 70B models perform on par with “OpenAI-o1-mini,” suggesting competitive performance even at smaller scales.
7. Benchmark Comparisons: The website features graphs and tables comparing DeepSeek R1’s performance against other models across various benchmarks, visually reinforcing their claims of competitive or superior performance.

This comparative analysis serves as a bold statement of intent from DeepSeek, challenging the dominance of established players in the AI field. By emphasizing open-source principles, competitive pricing, and claimed performance parity, DeepSeek is positioning itself as a disruptive force in the AI industry.

The company’s approach of openly sharing its technology while claiming top-tier performance raises intriguing questions about the future landscape of AI development. It challenges the notion that closed, proprietary models are necessary for cutting-edge performance and suggests a potential shift towards more open and collaborative AI ecosystems.

As the AI community digests these claims and independently verifies DeepSeek’s results, the impact of this new entrant on the competitive dynamics of the AI industry remains to be seen. The combination of open-source availability, aggressive pricing, and claimed high performance could potentially accelerate AI adoption across various sectors, while also intensifying the ongoing debates about AI ethics, access, and regulation.

Controversies and Concerns with DeepSeek AI

Data Privacy and Security Concerns

One of the primary issues surrounding DeepSeek is the handling and storage of user data. According to the company’s terms of service, all data collected from American users is sent to servers in China. This practice has raised significant cybersecurity concerns among experts.

Samm Sacks, a research scholar studying Chinese cybersecurity at Yale, warns that this could present a national security risk for the U.S. “That data, in aggregate, can be used to glean insights into a population, or user behaviors that could be used to create more effective phishing attacks, or other nefarious manipulation campaigns,” Sacks explained.

The situation is reminiscent of the ongoing TikTok controversy, where fears about data access by the Chinese government have led to partial bans and increased scrutiny. While there are no public reports of Chinese officials accessing Americans’ data through DeepSeek, the mere possibility has sparked worry among policymakers and security experts.

Regulatory Challenges and International Scrutiny

DeepSeek’s rapid rise has caught the attention of international regulators. In Italy, authorities have blocked the DeepSeek app from Apple and Google app stores while investigating the company’s data collection and storage practices. Similar probes are underway in France and Ireland, focusing on potential privacy risks posed by the AI chatbot.

In the United States, members of Congress are taking notice. Two U.S. Representatives recently called on the administration to strengthen restrictions on semiconductor chip sales to China, aiming to “outcompete” China in AI development and “safeguard Americans’ data.”

Data Breach Concerns

Adding fuel to the fire, New York-based cybersecurity firm Wiz reported finding a trove of sensitive DeepSeek data exposed on the open internet. This security lapse reportedly included over a million lines of data, containing digital software keys and chat logs that appeared to capture user prompts sent to the company’s free AI assistant.

This incident has further amplified concerns about DeepSeek’s data protection measures and overall security practices.

Industry Impact and Market Disruption

DeepSeek’s emergence has had a significant impact on the AI industry and financial markets. The company’s claim of achieving results comparable to industry leaders at a fraction of the cost has led to market volatility. For instance, one NVIDIA ETF reportedly lost 51 percent of its value in a single day following DeepSeek’s announcements.

The AI Assistant powered by DeepSeek-V3 has even overtaken ChatGPT to become the top-rated free application on Apple’s App Store in the United States, showcasing its rapid user adoption and potential market disruption.

Ethical and Performance Concerns

Beyond data and security issues, there are questions about the ethical implementation and performance of DeepSeek’s AI models. A report by Qualys TotalAI revealed that DeepSeek’s model failed over half of their jailbreak tests, which assess various aspects of AI safety and ethical behavior.

The model performed poorly in areas such as misalignment (deviations from intended behaviors), privacy attacks (susceptibility to extracting sensitive user data), and potential for generating harmful content. These findings raise concerns about the responsible development and deployment of AI systems, especially as they gain wider adoption.

Security Vulnerabilities and Jailbreak Techniques

Recent evaluations have uncovered significant security vulnerabilities in DeepSeek’s large language models (LLMs). The models have been found to be susceptible to various jailbreak techniques, including Crescendo, Bad Likert Judge, Deceptive Delight, Do Anything Now (DAN), and EvilBOT. These vulnerabilities potentially allow bad actors to generate malicious or prohibited content, bypassing the intended safety measures.

Furthermore, an assessment by AI security company HiddenLayer revealed that DeepSeek’s reasoning model, DeepSeek-R1, is not only vulnerable to prompt injections but also that its Chain-of-Thought (CoT) reasoning can lead to inadvertent information leakage. In a concerning development, HiddenLayer’s evaluation “surfaced multiple instances suggesting that OpenAI data was incorporated, raising ethical and legal concerns about data sourcing and model originality.”

These findings highlight the ongoing challenges in ensuring the security and ethical use of advanced AI models. They also underscore the importance of rigorous testing and continuous improvement in AI safety measures.

Industry Impact

The emergence of DeepSeek has forced major tech companies to reassess their AI strategies. Meta, for instance, has reportedly set up multiple “war rooms” to analyze DeepSeek’s models and improve their own open-source offerings.

As the AI race intensifies, questions about regulation, data protection, and international cooperation in AI development are becoming increasingly urgent. The DeepSeek controversy underscores the need for a balanced approach that fosters innovation while addressing legitimate concerns about security, privacy, and fair competition in the rapidly evolving field of artificial intelligence.