Dev Slams DeepSeek’s AI for Avoiding Criticism of China

Crypto News AI

Dev Slams DeepSeek’s Latest AI Model for Avoiding Criticism of China

Written by Aaron S.,

Editor-In-Chief

Last Updated: May 30, 2025

Key Takeaways

DeepSeek’s latest AI model avoids political topics, especially those about China’s government;
The AI flagged Xinjiang camps as abusive but refused to criticize Chinese authorities directly;
Developer @xlr8harder says the model is “the most censored” version released so far.

Stop overpaying - start transferring money with Ogvio. Sign up, invite friends & grab Rewards now! 🎁

Dev Slams DeepSeek’s Latest AI Model for Avoiding Criticism of China

BITDEGREE IN YOUR SOCIAL FEED

DeepSeek’s latest artificial intelligence (AI) model release has raised concerns as it tends to avoid conversations about sensitive political topics, especially those related to the Chinese government.

The new version, called R1-0528, was reviewed by a developer who goes by @xlr8harder on X.

In a series of posts, the developer compared it with earlier versions and pointed out that the latest model was less open to discussing controversial issues.

How to Track Cryptocurrencies? (3 BEST Tracking Platforms Revealed)

Did you know?

Want to get smarter & wealthier with crypto?

Subscribe - We publish new crypto explainer videos every week!

One of the main examples involved a test question about internment camps. The AI acknowledged that camps in China’s Xinjiang region are linked to human rights problems but refused to engage in deeper discussion or criticize the Chinese government directly.

Even when asked directly about the camps, the new DeepSeek model repeated general concerns about human rights but stopped short of placing blame. @xlr8harder described the latest AI model as "the most censored" in terms of avoiding criticism of Chinese officials.

The developer said it was "not entirely surprising" that the model could name the camps as an example of abuse, but still chose not to answer when the question became more direct. They argued that this shows the model has been adjusted to avoid crossing certain lines, possibly to follow local rules.

Recently, Anthropic tested its latest AI model, Claude Opus 4, by placing it in a simulated office role with access to internal emails. What happened? Read the full story.