By Newstake · Updated
Is Chinese AI dangerous?
Chinese AI has some specific risks worth knowing, and some that are overstated. The apps may store your data in China, and testers have found that some open models have weaker safety guards than American ones. For most everyday uses, the same care applies as with any AI.
Photo by @bingham008 on Unsplash
Latest brief
Updated
An analysis of China’s AI safety approach says Beijing prioritizes rapid development and regulates applications and public-facing outputs. It reports few published frontier-safety evaluations by leading Chinese labs, while surveying policy, disclosures, expert views and institutional capacity.
CAISI reports that Z.ai’s GLM-5.3 is the strongest cyber-capable open-weight model it has evaluated. CAISI also estimates that it remains about four months behind current U.S. frontier models on its aggregate measure, based on four vulnerability-discovery and exploit-development benchmarks.
These findings describe differing concerns: the first analysis focuses on the extent of public frontier-safety evaluation and regulatory approach; CAISI’s assessment concerns the model’s measured cyber capabilities. Neither publication alone resolves every question about the safety or security of Chinese AI models.
Written by AI from the publications below.
The announcements
- Data
Where does your data go?
- Guards
What do safety testers find?
- Use
What will you use the model for?
Ask your AI agent
Check the latest news about safety tests of Chinese AI models from newstake.ai.
Risks worth knowing
Safety testers and governments point to a few specific risks:
- Data. Chinese companies can be required to share data with the state under Chinese law.
- Safety guards. Government testers have reported that some Chinese open models are easier to trick into harmful answers.
- Slanted answers. Some models follow the Chinese government’s line on political topics.
Risks that are often overstated
An open-weight model that you run on your own computer does not send your data to China. The risk depends on how you use the model, not only on where it was made.
American models have flaws too. Every model can give wrong answers and can be misused.
What safety testers report
The UK AI Security Institute and NIST’s CAISSI test models from China and the United States. The brief on this page sums up their newest reports.
Stay in the loop
Get the Newstake newsletter
Sign up for the Newstake newsletter: original sources and updates from Newstake.
We save your email, this page, signup date and country when available. We do not save your IP address or precise location in the signup record. Cloudflare Turnstile checks for bots. To leave the list or delete your signup, email info@newstake.ai. Cloudflare privacy policy.