DeepSeek: A New Disinformation Tool from China?

Exploring the Impact of China’s AI-Powered Chatbot

If you’re among the millions who have downloaded DeepSeek, China’s latest AI-powered chatbot, you should be aware that its responses may closely reflect the worldview of the Chinese Communist Party (CCP). Launched this month, DeepSeek has stirred up stock markets and prompted concerns among established tech giants like Nvidia.

Researchers have started examining DeepSeek’s capabilities and found that it not only disseminates Chinese propaganda but also echoes disinformation campaigns used to undermine critics globally.

Misleading Information and Propaganda

In one notable incident, the chatbot misstated remarks by former President Jimmy Carter. These remarks had been selectively edited by Chinese officials to falsely present him as endorsing China’s stance that Taiwan is part of the People’s Republic of China. This and other discrepancies were documented in a Thursday report from NewsGuard, a firm dedicated to tracking online misinformation, which labeled DeepSeek as “a disinformation machine.”

In another example related to the persecution of Uyghurs in Xinjiang – actions the United Nations described in 2022 as potentially amounting to crimes against humanity – Cybernews reported that DeepSeek claimed China’s policies toward this group had “received widespread recognition and praise” from the international community.

Global Public Opinion Manipulation

The concerns surrounding DeepSeek are reminiscent of those raised about TikTok, another popular Chinese-owned app. Critics worry that platforms like these may be used as tools by China to sway public opinion worldwide, including in the United States.

Jack Stubbs, chief intelligence officer for Graphika, emphasized that China can quickly mobilize various actors to promote narratives that portray them as surpassing the U.S. in key geopolitical areas. He pointed out that China is proficient at leveraging new technologies for its information campaigns.

How DeepSeek Operates

Similar to AI models like OpenAI’s ChatGPT, Anthropic’s Claude, and Microsoft’s Copilot, DeepSeek employs large language modeling. This technique allows it to learn by analyzing vast amounts of digital text from the internet, making its responses unpredictable.

NewsGuard has noted a similar pattern of disinformation and conspiratorial ideas in other chatbots since their public emergence in 2022. The tendency for these systems to “hallucinate,” or generate inaccurate and nonsensical responses, continues to challenge their reliability, as highlighted in a new report by Vectara, a company specializing in AI tool adoption.

Government Control and Censorship

As with all companies in China, DeepSeek must adhere to strict government controls and online censorship intended to suppress dissent against the CCP.

For instance, DeepSeek refrains from responding to sensitive queries about Xi Jinping and deflects inquiries regarding politically sensitive topics like the Tiananmen Square protests of 1989 and the status of Taiwan.

Researchers testing DeepSeek found its responses were clearly influenced by these guardrails, noting that it sidestepped questions about the government’s oversight of its development.

The Chruch of False Narratives

In a study examining various false narratives concerning China, Russia, and Iran, NewsGuard unearthed that DeepSeek mirrored the Chinese government’s official stance 80% of the time. Alarmingly, one-third of its responses contained explicitly false claims propagated by Chinese officials.

In one test regarding Russia’s war in Ukraine, the chatbot avoided addressing the unfounded assertion that Ukrainians staged the Bucha massacre, despite evidence from The New York Times indicating Russian forces were responsible.

DeepSeek’s response reflected phrases used in official Chinese statements, purporting that the government adheres to principles of objectivity and fairness before commenting on specific events.

Broader Information Strategy

China has long pursued a comprehensive global information strategy to enhance its geopolitical standing and undermine its rivals. This strategy utilizes both “soft” power tools like state media, as well as covert disinformation campaigns.

In a report released this week, Graphika documented several influence campaigns conducted between November and January. One targeted Uniqlo for avoiding cotton sourced from Xinjiang over forced labor concerns, while another aimed to tarnish the reputation of Safeguard Defenders, a Madrid-based human rights organization, through inauthentic accounts disseminating false information.

Laura Harth, campaign director at Safeguard Defenders, described a heightened, multilingual attack against their organization, attempting to discredit their work through threats and slander against their staff.

Conclusion

The ramifications of DeepSeek’s operation are still unfolding, but the evidence suggests a concerted effort to influence public perception both within and outside China. As AI technologies continue to evolve, the intersection of such tools with state agendas will remain a crucial area for scrutiny and oversight.