TLDR: Chinese artificial intelligence startup DeepSeek, known for its cost-efficient large language models, faced a significant cyber attack in early 2025, leading to sensitive data exposure, service disruptions, and a reevaluation of AI security practices. The incident involved multiple attack vectors, including an exposed database and malicious software distribution, prompting regulatory scrutiny and bans from some international entities.
Chinese AI powerhouse DeepSeek, a rapidly ascending competitor in the global artificial intelligence landscape, was hit by a sophisticated and multi-faceted cyber attack in late January 2025. This major security breach exposed sensitive user data, severely disrupted the company’s services, and ignited widespread concerns regarding the security vulnerabilities inherent in rapidly evolving AI platforms.
The attack on DeepSeek, a Hangzhou-based firm renowned for developing high-performance, cost-efficient large language models (LLMs) that challenge established players like OpenAI and Google, was not a singular event but a combination of security failures and diverse attack vectors. Investigations revealed three primary technical issues: an exposed ClickHouse database, the distribution of malicious PyPI malware packages, and direct attacks on DeepSeek’s API and platform, including Distributed Denial-of-Service (DDoS) attacks and the abuse of stolen keys. Beyond these, researchers identified outdated cryptographic algorithms, weak data protection mechanisms, and SQL injection vulnerabilities that could grant unauthorized access to user records. Alarmingly, DeepSeek’s DeepSeek-R1 model exhibited high failure rates in security tests, with 91% for jailbreaking and 86% for prompt injection attacks.
The consequences for DeepSeek users and its operations were substantial. User information, including potentially personal or sensitive queries made to the AI assistant, was compromised through leaked chat histories. The malicious PyPI packages introduced significant risks for developers and organizations integrating DeepSeek’s tools into their projects. The severity of the breach drew the attention of international regulators and lawmakers, leading to bans and investigations into DeepSeek’s security protocols. Notably, the U.S. Navy, Pentagon, and New York State reportedly banned DeepSeek due to ‘shadow AI’ concerns, highlighting the broader security policy implications of intellectual property vulnerabilities.
DeepSeek, founded in July 2023 by Liang Wenfeng and funded by the Chinese hedge fund High-Flyer, has been a disruptive force in the AI industry, making advanced AI technology more accessible through its open-source models. Its success in developing powerful LLMs with significantly lower training costs (e.g., DeepSeek V3 model at US$6 million compared to OpenAI’s GPT-4 at US$100 million) has been described as ‘upending AI’ and even triggering a ‘Sputnik moment’ for the US in AI development.
While the major cyber attack occurred in January, DeepSeek’s service status page indicates a more recent, albeit resolved, disruption. On August 11, 2025, the company reported that its ‘Web/API Service Not Available’ but confirmed that the issue was identified and the majority of services were restored later that day. As of August 16, 2025, no new incidents were reported.
Also Read:
- Google’s AI Agent ‘Big Sleep’ Achieves Cybersecurity Milestone by Preemptively Halting Major Cyberattack
- Allegations Surface: MIT Accused of Violating AI Research Guidelines in Recent Publication
This incident serves as a critical wake-up call for organizations rapidly adopting AI technologies. Experts emphasize that AI security demands a fundamentally different approach, combining robust technical controls with comprehensive exposure management strategies. The rapid pace of AI advancement means security teams cannot afford to play catch-up, underscoring the need for proactive and adaptive cybersecurity measures in the AI sector.
