Chinese AI model DeepSeek raises deep concerns about propaganda

DeepSeek‘s release highlights growing concerns about how AI models trained with cultural or political biases could be weaponized for propaganda purposes. While much of the debate around this Chinese-made large language model has focused on cybersecurity and intellectual property concerns, the potentially more significant threat lies in how such models—designed as training tools for future AI systems—could be used to shape global narratives and spread state-approved worldviews across international borders.

The big picture: DeepSeek’s design as a foundation model for training other AI systems raises concerns about embedded political biases being propagated through future technology.

• The Chinese AI model was explicitly unveiled in January as “an encyclopedia to train the next generation of AI models,” positioning it as an influential knowledge source.
• Organizations building smaller, specialized AI systems may inadvertently incorporate biases from foundation models like DeepSeek, potentially spreading state-approved narratives.

Why this matters: As AI development costs decrease, the barrier to creating models with embedded cultural and political perspectives has lowered significantly.

• The proliferation of politically-influenced AI systems could become a new frontier in global information warfare and soft power competition.
• AI bias isn’t a new concern, but previous examples have often appeared unintentional rather than strategically engineered.

What they’re saying: Industry experts warn about AI’s potential to become a geopolitical influence tool.

• “AI models have an opportunity to shape the narrative, shape the minds, shape the outcomes,” notes Dhaval Moogimane of West Monroe.
• Adnan Masood from UST describes AI as becoming “a force multiplier for ideological compliance and national soft-power export.”

Broader context: DeepSeek isn’t unique in raising these concerns about AI’s potential for propaganda.

• Experts acknowledge that many large language models will likely serve as training tools for future systems.
• Multiple governments and organizations—not just China—could leverage AI models as vehicles for propaganda and narrative control.

Recommended safeguards: Organizations can take several steps to mitigate AI bias and propaganda risks.

• Creating modular AI architectures allows for more transparency and control over system components.
• Implementing continuous monitoring of AI tools can help identify emerging biases or problematic outputs.
• Utilizing established frameworks like the NIST AI Risk Management Framework provides structured guidance for responsible AI deployment.