China's new AI can talk politics, but only in code
Huawei and Zhejiang University have released DeepSeek-R1-Safe, a fresh take on the DeepSeek language model.
What's different? This version is carefully programmed to steer clear of politically sensitive topics, matching strict government guidelines.
It was retrained using 1,000 Huawei AI chips—so it avoids nearly all flagged content in ordinary interactions but still retains about 99% of its original capability and speed.
The AI landscape is becoming increasingly polarized
DeepSeek-R1-Safe isn't alone—countries are shaping AI to fit their own values.
Earlier this year, Saudi Arabia launched a chatbot focused on Islamic culture, while in the US, new rules require government AIs to stay neutral on hot-button issues.
The takeaway: more and more, where you build your AI shapes how it thinks and what it talks about.