Can we trust Chinese AI? I considered asking Kimi K3 for its opinion. Then I realised the obvious problem with that plan.
In Part 1 I wrote about being entrenched in Claude Co-Work. In Part 2 I broke down why Chinese AI models are so much cheaper. Today I want to talk about the part nobody covers. The trust question. And the extraction nightmare.
What open-weight actually means
Kimi K3 is an open-weight model. That means you can download it. Run it on your own hardware. Modify it. No permission needed. No server in Beijing involved.
This is different from using an API where your data flows through someone else’s servers. With an open-weight model running locally, your data stays on your machine. That sounds reassuring until you realise the Trump administration is still trying to ban Chinese AI models entirely.
The concern isn’t just about data flowing to China. It’s about the model itself. What if the model has hidden behaviours? What if it’s been trained to subtly influence outputs in ways that benefit Chinese interests? Nobody has proven this. But nobody has disproven it either.
The ban that can’t be enforced
The Trump administration is reportedly reviving a push to ban Chinese AI models, citing cybersecurity concerns. Treasury Secretary Scott Bessent has warned that Chinese AI firms could face sanctions if there’s evidence their models rip off US models.
The problem is enforcement. You can’t ban an open-weight model. It’s already downloaded. It’s on servers across the US. It’s being used by developers right now. Banning it is like trying to ban an MP3 file. The technology doesn’t care about borders.
Meanwhile, Xi Jinping used the World Artificial Intelligence Conference in Shanghai last week to call for countries to oppose the practice of overstretching national security in AI. He launched a new AI alliance. He positioned China as the champion of open-source AI. It was a direct shot at Washington’s controls.
Should you be worried?
Honestly? I don’t know. And I’m not going to pretend I do.
What I do know is this. Every US tech company I’ve trusted has had data breaches. Every single one. Google, Meta, Microsoft, Amazon. All of them have leaked user data at some point. The idea that American AI companies are inherently more trustworthy because they’re American is a story we tell ourselves, not a fact.
But China has different data laws. Very different. The Chinese government can compel Chinese companies to hand over data. US companies face similar pressures through Patriot Act provisions, but the Chinese system is more opaque. There’s less daylight between the state and the corporation.
If you’re running a small business and feeding client data through a Chinese AI model, you need to think about this. Not panic. Think.
The extraction nightmare
But let’s say you decide to switch. You’re comfortable enough with the security question. Kimi K3 is a third of the price. Let’s go.
How do you actually do it?
I’m going to use myself as a cautionary tale. When I moved from ChatGPT to Claude Co-Work, I requested a data export from ChatGPT. All my conversation history. All my custom instructions. Everything.
I requested it twice. I never received anything. Not an email. Not a download link. Not even an acknowledgement. My entire ChatGPT history is just gone. Trapped in OpenAI’s servers with no way out.
Now imagine extracting from Claude Co-Work. It’s not just chat history. It’s a desktop app connected to my creative workflow. Image optimisation. Social media content design. Recurring scheduled tasks. Domain name management. All non-sensitive stuff, but it’s still months of setup and configuration.
You can’t just export a settings file. You’d need to rebuild every workflow. Reconnect every tool. Recreate every scheduled task. Re-train the new model on your preferences. It’s not a switch. It’s a migration project.
And the AI companies know it. The harder it is to leave, the less likely you are to try. Every integration you build, every workflow you automate, every scheduled task you set up, it’s another tentacle holding you in place.
How to test without committing
If you’re curious about Kimi K3 but not ready to burn everything down, this is what I’d suggest.
Try it on kimi.com. It’s free. Ask it the same questions you’d ask your current AI. See if the output quality holds up for your specific use case. Not benchmarks. Your actual work.
Run a side-by-side test for a week. Same prompts, both models. Compare the output. Compare the cost. See if the quality gap justifies the price gap.
If it passes, look at running the open-weight version locally. That keeps your data on your hardware. No Chinese servers in the loop. You get the pricing benefit without the data flow concern.
And if it doesn’t pass? You’ve lost nothing but a few hours. And you’ve confirmed your current setup is actually the right one. That’s worth knowing too.
My honest verdict
I haven’t switched. I’m not sure I will. The pricing gap is real and significant. But the extraction cost is real too. And the trust question doesn’t have a clean answer.
What I will do is keep testing. Keep watching. Keep asking uncomfortable questions. Because the AI landscape is shifting under our feet, and the companies that win won’t be the ones with the best marketing. They’ll be the ones that earn trust and make it easy to leave.
Right now, nobody earns full trust. And nobody makes it easy to leave.
The China Question series
Part 1 — I’m So Deep in Claude Co-Work I Can’t See the Surface
Part 2 — A Third of the Price. What’s the Catch with Chinese AI?
Frequently Asked Questions
Can the US government actually ban Chinese AI models?
An outright ban is nearly impossible to enforce because most Chinese models are open-weight. Once downloaded, they can run on any hardware without connecting to Chinese servers.
How do I test Kimi K3 without switching from my current AI?
Visit kimi.com for free access, or use the API to run side-by-side comparisons with your current model. Run the same prompts through both and compare quality and cost.
What happens to my data if I switch to a Chinese AI model?
If you use the API, data flows through the provider’s servers. If you run the open-weight version locally, data stays on your hardware. Always check the provider’s data policy before feeding sensitive business or client information through any AI model.
Written by a human. Affiliate links may apply. Nuffin’s for free.
