Microsoft Prohibits Employees From Using ‘Chinese Propaganda’ Chatbot

Microsoft Bans Employee Use of Deepseek

Microsoft has prohibited its employees from using the Deepseek application, which has raised concerns as a potential source of “propaganda,” according to President Brad Smith. This announcement was made during a Senate testimony where he explained that the app has been blocked on all Microsoft devices and removed from the Windows Store. Engineers were worried that it could leak sensitive employee data to servers in China while promoting pro-CCP narratives.

Smith’s statements come amid broader efforts to discourage the use of Deepseek within the U.S., particularly in government and corporate sectors. He mentioned, “Microsoft does not allow employees to use the Deepseek app,” due to fears regarding data being returned to China and possibly influencing responses with propaganda.

In January, an engineer’s findings about Deepseek gained significant attention. Smith noted he examined the open-source R1 model of Deepseek to modify the code and mitigate any harmful effects. The modified version was then stress-tested and made available on Azure’s cloud marketplace. Deepseek has not responded to requests for comments regarding these developments.

The recent Senate hearing underscored concerns in Washington about AI tools from Beijing that circumvent U.S. export controls while harvesting American data. Deepseek surged into the top ranks of apps on both Apple and Google platforms, offering ChatGPT Plus-level capabilities for free. Meanwhile, Deepseek explicitly stated that it would collect and store personal data within China.

Microsoft’s actions, which extend beyond merely criticizing competitor OpenAI’s Deepseek, align with a growing trend in the public sector. Bipartisan lawmakers have introduced legislation to restrict the use of such tools on government devices.

Nonetheless, Microsoft’s policies aren’t exhaustive. While direct usage of the Deepseek app is not permitted, the R1 model remains accessible for customers who want to customize it within their networks. Smith assured that this approach could be safely managed with sufficient internal controls.

He elaborated, “Deepseek produced two things: an open-source model and the Deepseek app. We don’t allow the app, primarily due to concerns over data potentially being sent back to China, which could encourage propaganda-related outputs.” Microsoft actively analyzes and modifies such models to eliminate any harmful impacts.

The rise of Deepseek in the U.S. has also sparked fears regarding intellectual property theft. Following an incident involving API scraping last fall, security teams from Microsoft and OpenAI have scrutinized whether Deepseek illegally extracted elements from their GPT models.

An OpenAI representative mentioned that companies based in the People’s Republic of China are known to replicate leading U.S. AI models. To address these threats, precautions are being taken to protect intellectual property and to carefully manage how new features are introduced in their models.

As discussions on how Congress might counter China’s rapid advancements in AI continue, Smith emphasized the need for significant American computing power and a balanced federal approach towards foreign platforms that might act as surveillance conduits.