DeepSeek AI is a cutting-edge AI model with groundbreaking capabilities and significant data privacy and security concerns. Organisations planning to deploy such advanced AI models in real-world applications need to be aware of the security vulnerabilities and regulatory challenges they could face.

The state of data privacy in DeepSeek

The issue of data protection with DeepSeek AI is pressing and demands immediate attention as the potential privacy concerns could have serious repercussions.

While providing free, robust AI services, DeepSeek may collect various data, such as detailed user interactions, location data, and even biometric information. This mass data accumulation makes users vulnerable to unauthorised surveillance and data breaches and opens the door to potential misuse by third parties.

For example, under China’s cybersecurity law, authorities can request access to data from any company operating within its jurisdiction, raising serious privacy concerns for global users.

DeepSeek models can be designed to repurpose the collected data due to non-enforcement of strict safeguards. This could lead to unauthorised profiling, mass surveillance, or even data leaks, especially if the model is subject to oversight from entities with unclear or opaque data governance policies.

These risks would be amplified if DeepSeek were to be integrated into third-party applications that continuously extract and share user data without explicit, transparent disclosure.

The potential misuse of user data is a serious concern, especially if the DeepSeek platform shares user data across borders, risking exposing sensitive information to countries with less stringent privacy regulations, increasing the potential for misuse leading to data leaks and regulatory violations under laws like GDPR and CCPA by circumventing the regulatory requirements for data minimisation.

Users interacting with the DeepSeek platform may not be fully aware of what data is retained, how long it is stored, or whether it is used for model retraining (updating the AI model with new data sets to improve its performance), behavioural analysis or external data-sharing agreements. This lack of awareness impacts transparency and explainability provided by DeepSeek AI models.

Security pitfalls in DeepSeek

As well as data privacy concerns, the emergence of DeepSeek AI introduces potential security threats, particularly concerning its abuse, surveillance risks and model exploitation – the malicious use of AI models to achieve unauthorised objectives.

As the security posture of the DeepSeek AI service is not clearly defined, there may be inherent vulnerabilities such as encryption flaws, hardcoded keys in mobile applications, and injections that could be exploited by attackers, leading to unauthorised operations and the potential exfiltration of user-centric data.

Furthermore, these vulnerabilities could enable the successful execution of reverse engineering of the DeepSeek AI model through inference attacks to decipher the internal workings of the model, potentially leading to the manipulation of the AI model. Adversaries can also execute membership inference attacks that allow attackers to verify the presence of specific data objects in training data sets.

DeepSeek may reinforce hallucinations and biases in training data, leading to inaccurate, misleading or manipulated content. If AI operators or attackers manipulate DeepSeek’s training data through data poisoning, it enhances the risk of automated disinformation campaigns, where adversaries exploit AI-generated content maliciously. This compromises AI integrity and can be used to spread misinformation or launch targeted cyber-attacks.

DeepSeek operates in different geographical locations on centralised cloud services, where obtaining complete visibility is not viable. This makes it susceptible to supply chain attacks, insider threats and unauthorised access. The lack of transparency in AI infrastructure is a significant concern, raising questions about security backdoors or embedded vulnerabilities.

Nation-state influence and industrial espionage are also significant threats. DeepSeek AI’s potential ties to state-sponsored entities introduce risks of data exfiltration, mass surveillance and corporate espionage. The AI insights generated by DeepSeek could be weaponised for misinformation, political influence or cyber warfare.

DeepSeek is data- and compute-intensive, making it a prime target for distributed denial of service (DDoS) attacks. Cybercriminals can overwhelm DeepSeek’s servers by flooding them with excessive requests, disrupting services and causing downtime. If the inherent APIs and associated infrastructure fail to handle application-layer DDoS attacks triggered by botnets, it could impact the availability of the DeepSeek AI service to many users.

Lessons for organisations from DeepSeek

DeepSeek AI is just one example of how unregulated AI applications can introduce security and privacy risks. For organisations integrating AI-powered applications, defending against AI abuse and cyberattacks must be a top priority.

Organisations must adopt transparent data collection policies and implement privacy-by-design principles. Providing users with precise opt-in mechanisms is crucial to ensure ethical and secure AI deployment.

Aditya K Sood VP of Security Engineering and AI Strategy Aryaka