Governments in the U.S, Italy, and now Australia have moved to ban access to DeepSeek, a Chinese-developed large language model (LLM), on government devices. These bans stem from concerns over data security and the potential exposure of sensitive information to the PRC.

However, blocking access to DeepSeek is not as simple as restricting its website and mobile app—because the model is open-source, it can still be run locally, through cloud environments, or integrated into various applications.

Satnam Narang, Senior Staff Research Engineer, Tenable

Various bans have been placed on DeepSeek, a large language model (LLM) developed by a Chinese start-up, on government devices around the world including in the United States (state of Texas), Italy, and now Australia. These actions are being taken due to the risks, specifically around usage of DeepSeek’s website and mobile app due to fears of exposure of sensitive information to the People’s Republic of China (PRC).

Banning access to the DeepSeek website and mobile applications is straightforward. However, because the DeepSeek LLM itself is open-source, blocking access to it is less straightforward, as it can be run locally on a device, it can be run through a cloud environment, and installed via various apps or frameworks like ollama. It requires some extra leg-work for users to set-up, but it can be done.

The local LLMs don’t require access to external servers, minimising the risk of leaking sensitive data. However, banning DeepSeek’s website and apps on government devices is one barrier, because they are still accessible on personal devices, which could be used to share sensitive information. Another concerning aspect to DeepSeek is not just the exposure of sensitive information, but rather the model’s lack of safety features, as it could be used for harmful or nefarious purposes. Satnam Narang, Senior Staff Research Engineer, Tenable.