In a recent Senate hearing, Microsoft's vice chairman and president, Brad Smith, revealed that employees at Microsoft are prohibited from using the DeepSeek app due to concerns about data security and potential influence from Chinese propaganda. This marks the first public acknowledgment by Microsoft regarding such restrictions on DeepSeek usage. Despite these reservations, Microsoft has made DeepSeek’s R1 model available through its Azure cloud service. However, this differs from offering the chatbot app itself, as users can download and host the open-source model independently without sending data back to China. Nonetheless, risks like the spread of propaganda or insecure code persist.
Details of Microsoft's Stance on DeepSeek
During a significant Senate hearing held recently, in a room filled with policymakers and tech industry observers, Brad Smith, Microsoft's high-ranking official, addressed pressing issues concerning artificial intelligence applications. He stated clearly that within Microsoft's corporate environment, employees are barred from engaging with the DeepSeek application, which is accessible both on desktops and mobile devices. This decision stems from worries that user data could be stored on servers located in China, thus being subject to Chinese legal mandates requiring collaboration with intelligence agencies. Moreover, DeepSeek enforces stringent censorship on topics deemed sensitive by the Chinese government.
Smith highlighted that these concerns also led Microsoft to exclude DeepSeek from its app store offerings. While numerous organizations and even nations have instituted limitations on DeepSeek, this represents the initial instance where Microsoft has openly declared such a prohibition. The rationale behind this restriction involves apprehensions over data storage locations and the possibility of responses being swayed by governmental propaganda. Despite these critical remarks, shortly after DeepSeek gained widespread attention earlier this year, Microsoft introduced DeepSeek’s R1 model on its Azure cloud platform.
It's important to note that making DeepSeek’s R1 model available via Azure differs from providing the chatbot app directly. Given that DeepSeek operates under an open-source framework, any entity can procure the model, host it on private servers, and offer it to clients without transmitting data back to China. Nevertheless, other hazards such as the dissemination of biased information or the generation of unsafe code remain. During the Senate session, Smith mentioned that Microsoft had managed to delve into DeepSeek’s AI model and modify it to eliminate detrimental side effects. Microsoft did not specify the exact modifications made to DeepSeek’s model, referring inquiries to Smith's testimony.
At the outset of launching DeepSeek on Azure, Microsoft indicated that the model underwent thorough evaluations and safety assessments before being deployed on Azure. It's worth noting that while DeepSeek competes directly with Microsoft's Copilot internet search chat app, not all competing chat platforms face bans from Microsoft's Windows app store. For example, Perplexity remains accessible in the Windows app store. Conversely, apps from Microsoft's major competitor, Google, including the Chrome browser and Google's chatbot Gemini, do not appear in searches within Microsoft's webstore.
From a journalistic perspective, this situation underscores the intricate balance tech giants must navigate between fostering innovation and ensuring data security. It raises questions about the implications of geopolitical influences on technological development and the measures companies take to mitigate potential risks. As artificial intelligence continues to evolve, so too must the ethical frameworks governing its use, prompting reflection on how best to protect user data while promoting technological advancement.
