Chinese open-weight AI models spread abroad, raising propaganda concerns

In short: Chinese open-weight AI models are being widely adopted and customized outside China, and experts warn this could later be used for online influence and propaganda.

What's going on

Chinese “open-weight” AI models are increasingly being used by foreign governments and companies. Open-weight means the model’s core settings are shared, so others can download it and modify it, like getting a recipe you can tweak instead of only ordering the dish.

Recent examples include Singapore’s national AI effort moving from Meta’s Llama model to Alibaba’s Qwen3-32B, and Malaysia’s communications ministry launching a national AI setup based on DeepSeek. Reporting also describes companies and startups across Southeast Asia, Africa, the UAE, and Japan building products on models like DeepSeek and Qwen for tasks such as chatbots, writing summaries, coding help, and local-language tools.

Some projects combine a local version of the model with “AI agents.” An AI agent is software that can take steps on its own, such as browsing the web, posting, and replying, instead of only answering a question. Security services have warned that this matters for propaganda. Estonia’s 2026 intelligence report said DeepSeek could leave out security-related information and insert pro-China messages in answers, and it warned that deep integration into digital systems could be activated around elections.

What to watch

The strongest public evidence points to a broad pattern, not one single coordinated campaign. The risk is that once these models are built into government services and everyday apps, they could help mass-produce tailored political messages or run large numbers of fake accounts with less human effort.

Source: NYTimes