Background
The rise of DeepSeek is undoubtedly a milestone in the development of AI technology in China. As a representative AI enterprise, DeepSeek has not only made breakthrough progress in technological innovation and commercial application, but also demonstrated the outstanding strength and great potential of Chinese technology enterprises in the global AI competition.
However, as DeepSeek’s international influence continues to grow, its challenges have become more complex and severe. Since the release of DeepSeek-R1 model with performance comparable to OpenAI o1 official version and significantly reduced cost, DeepSeek has suffered a series of targeted cyber attacks, posing a serious threat to its daily operations. The motives and intentions behind the attack are inevitably questionable. This phenomenon not only highlights the important strategic position of DeepSeek in the global AI field, but also reflects the complexity and fierceness of the international AI technology game, which is worth attention and caution.
Attack Monitoring
(i) Attacks against the DeepSeek API interface
The Global Threat Hunting System of NSFOCUS Security Lab detected 3 waves of DDoS attacks targeting IP address 1.94.179.165 at 15:33:31 on January 25, 2025, 13:12:44 on January 26, 2025, and 18:09:45 on January 27, 2025 respectively (GMT+8), which was the address resolved by DeepSeek’s API interface (api.deepseek.com). The average attack duration was 35 minutes. Attack methods mainly include NTP reflection attack and Memcached reflection attack.
Based on the advanced MoE model, the DeepSeek API supports dialogue generation and completion, empowering developers to further develop applications. As of 19:00 on January 28, 2025, the DeepSeek API interface development platform is still unavailable.
(ii) Attacks on the DeepSeek chat system
The Global Threat Hunting System of NSFOCUS Security Lab detected 2 waves of DDoS attacks on IP address 60.204.193.34 at 23:21:31 on January 20, 2025 and 11:18:21 on January 25, 2025 (GMT+8), which was the address resolved by DeepSeek’s chat system interface (chat.deepseek.com) at that time. Attack duration averaged over 1 hour. Attack methods mainly include NTP reflection attack and SSDP reflection attack.
It is worth noting that on January 20, 2025, when the DeepSeek chat system was attacked, DeepSeek officially released DeepSeek-R1. Reinforcement learning technology was widely used in the post-training phase of DeepSeek-R1, which greatly improved the model reasoning capability with only a few labeled data. Performance comparable to OpenAI o1 Official for tasks such as mathematics, coding, and natural language reasoning.
Sustained Attack Activities
At 00:06 on January 28, 2025 (GMT+8), the technical team of DeepSeek issued an emergency announcement confirming that its online services had suffered large-scale malicious attacks. After the incident, the DeepSeek technical team synchronously switched the primary domain name resolution IP address and adjusted the domestic resolution IP address from 1.94.179.165 to 60.204.2.236.
After the resolving IP address of DeepSeek is switched, the attacker quickly adjusted the strategy and launched a new round of DDoS attacks on the main domain name www.deepseek.com, the API interface api.deepseek.com, and the chat system chat.deepseek.com at 13:12:56 and 15:01:44 on January 28, 2025 respectively. The average attack duration is more than 30 minutes, the attack methods are NTP reflection, CLDAP reflection, etc., and the attack characteristics are basically consistent with those of previous attacks. As of 19:00 on January 28, the API interface has not been fully restored.
The attacker shows great pertinence and high tactical literacy. After DeepSeek completes the resolution address switching, the attacker quickly adjusts the attack policy and launches a new round of DDoS attacks on the core service system of DeepSeek. From the selection of attack targets to the accurate grasping of timing, and then to the flexible control of attack intensity, the attacker shows extremely high professionalism in every attacking step. This highly coordinated and precise attack suggests that the incident was not accidental, but likely a well-planned and organized cyberattack executed by a professional team.
Attack Resources
Through further traceability analysis, the top three sources of attack infrastructure that have been monitored are the United States (20%), the United Kingdom (17%) and Australia (9%).
Overseas attackers launched coordinated attacks on the target through highly concealed technical means, further highlighting the complexity of this attack. The attacker used globally distributed server resources to try to cover up the identity and geographical location. This behavior not only seriously undermined the order of cyberspace, but also posed a major threat to global Internet security.
Summary
This large-scale cyber attack on DeepSeek is a well-planned and organized attack with clear objectives. The motive and purpose of attackers choosing to launch attacks at key nodes where DeepSeek has achieved remarkable results and attracted global attention is thought-provoking. This incident not only exposes the security risks of international cyberspace, but also reflects the technological competition and geopolitical game that Chinese technology enterprises may face in the process of globalization.
In the future, DeepSeek will not only need to make continuous breakthroughs in technological innovation but also strengthen its layout in network security protection to cope with multiple challenges that may arise in the process of globalization. AI enterprises must build a more complete security system to protect their own technological achievements and users’ rights and interests and contribute to the healthy development of global AI technology.
