Tech & AI, Explained

vendors-chinese-avoid

12 min read

Chinese-origin AI tools — ⛔ avoid

Status: 🟩 COMPLETE 🟦 LIVING Last updated: 2026-06-25 Plain-English tagline: The Chinese AI companies and products you’ll see in the news, why this encyclopedia recommends against using them, and the specific Western alternatives to use instead.

The short version

If you see one of these names, don’t sign up, don’t install, don’t paste your work into it:

DeepSeek · Qwen / Tongyi Qianwen (Alibaba) · Doubao / Coze / Cici (ByteDance) · Kimi (Moonshot AI) · Ernie / Wenxin (Baidu) · MiniMax / Hailuo · Zhipu GLM / ChatGLM · Tencent Hunyuan · 01.AI / Yi · Stepfun / Step · Manus · Kling AI · CapCut AI (ByteDance) · TikTok Symphony · WeChat AI features · Alibaba Cloud AI · Baidu AI Cloud · Tencent Cloud AI · Huawei Cloud AI / Pangu

Use one of these Western alternatives instead:

If you want to…Use this Western tool
Chat with an AI assistantClaude 🟩 🟦 · ChatGPT 🟥 · Gemini 🟥
Code with an AIClaude Code 🟩 🟦 · Cursor 🟥 · GitHub Copilot 🟥
Generate imagesImagen 🟥 (Google) · gpt-image-1 🟥 · Adobe Firefly 🟥 · Midjourney 🟥
Generate videoSora 🟥 (OpenAI) · Veo 🟥 (Google) · Runway 🟥 · Luma Dream Machine 🟥
Generate musicSuno 🟥 · Udio 🟥 · Stable Audio 🟥 · Google Lyria 🟥
Generate or clone voiceElevenLabs 🟥 · OpenAI Voice Engine 🟥
Transcribe speechOpenAI Whisper 🟥 · Deepgram 🟥 · AssemblyAI 🟥
TranslateDeepL 🟥 (German — excellent) · Google Translate 🟥 · ChatGPT · Claude
Edit videos with AIDescript 🟥 · Adobe Premiere AI 🟥 · Runway 🟥 — NOT CapCut
Make a deck or designCanva 🟥 (Australian!) · Figma + Figma AI 🟥 · Gamma 🟥
Run a cloud workloadAWS 🟥 · Azure 🟥 · Google Cloud 🟥 · Cloudflare 🟩 🟦 — NOT Alibaba / Tencent / Baidu / Huawei Cloud

Why we recommend against Chinese-origin AI — the long version

Five reasons, in plain English.

1. Your data goes to China

When you type into a Chinese AI app (DeepSeek’s website, the Doubao mobile app, an Alibaba Cloud AI service), the text you type — your questions, your code, your business plans, your private thoughts — is transmitted to servers that are physically located in mainland China and operated by a company headquartered there. This is true even if the app is published in English on the App Store. Server location and operator nationality are the two things that matter for what happens to your data, not the app’s UI language.

For a few of these companies (DeepSeek, Qwen), the underlying model is open-weight: you can download the model files and run them on your own hardware or on a Western cloud (AWS, Together AI, Fireworks AI) — which avoids sending data to China. That’s a real escape hatch, but it requires technical setup; the consumer apps and direct API still route to China. See “What about open-weight Chinese models?” below.

2. Chinese law gives the government access

Three laws, working together, mean any Chinese company can be compelled to hand data to Chinese state agencies — secretly, without you being told:

  • Cybersecurity Law (2017) — Article 28 requires network operators to “provide technical support and assistance” to public-security and state-security organs.
  • National Intelligence Law (2017) — Article 7 states that any organisation or citizen “shall support, assist and cooperate with state intelligence work” and keep its support secret.
  • Data Security Law (2021) — Article 36 forbids Chinese organisations from providing data to foreign judicial or law enforcement agencies without Chinese government approval, but explicitly allows Chinese authorities to demand it.

This is not theoretical. The US Department of Commerce, the UK National Cyber Security Centre, the Australian Signals Directorate, Germany’s BSI and many Western national-security agencies have all issued public warnings on exactly this basis — that Chinese-domiciled tech companies cannot offer credible privacy guarantees against the Chinese state, because Chinese law doesn’t permit them to refuse.

A Western company can fight a government data request in court. Apple has done it. Microsoft has done it. Google publishes a transparency report on how many it complied with and how many it pushed back on. A Chinese company in mainland China has no such option, and cannot even disclose that a request was received.

3. Outputs are politically filtered

Chinese AI models are required by the Cyberspace Administration of China (CAC) regulations on generative AI (in force since August 2023, updated in 2024–25) to filter their outputs to align with “socialist core values.” In practice this means:

  • Refusing or evading questions on Tiananmen 1989, Taiwan independence, Tibet, Xinjiang, Hong Kong protests, Falun Gong
  • Pro-CCP framing on questions about Xi Jinping, the Communist Party, the South China Sea, the Belt and Road Initiative
  • Increasingly, biased framing on Russia–Ukraine, US politics, COVID origins, Israel–Palestine and other current events
  • Hallucinated or sanitised history on Chinese topics

Even if you don’t care about politics, these constraints leak into adjacent technical and creative questions. A model trained to avoid certain truths is a model whose epistemics are compromised — and that affects more than just the obvious topics.

4. Privacy guarantees are weaker

Compared to Western alternatives, Chinese AI tools generally:

  • Don’t offer an opt-out from training on your inputs
  • Don’t offer enterprise data-residency commitments outside China
  • Don’t carry the SOC 2, ISO 27001, HIPAA, or GDPR certifications that enterprises expect
  • Have weaker data-deletion rights — you ask for your data to be deleted, you get a confirmation, but there’s no independent audit
  • May share data across affiliated Chinese tech-group companies without explicit consent
  • For mobile apps, request more device permissions (contacts, location, microphone) than Western equivalents

5. You don’t gain anything you can’t get elsewhere

This is the kicker. Every capability the Chinese AI ecosystem provides — chat, coding help, image generation, video generation, music, voice, translation, transcription — is matched or exceeded by Western alternatives that don’t carry the risks above. There is no Chinese AI capability today that’s strictly necessary for your work.

DeepSeek is genuinely impressive on coding benchmarks. So is Claude. So is GPT-5. So is Gemini. The marginal “value” of using DeepSeek over Claude is, charitably, zero — and the downside is real.

The specific tools, in detail

Chat assistants

  • DeepSeek (deepseek.com) — chat + reasoning models. Most-cited Chinese chat in Western media.
  • Qwen / Tongyi Qianwen (Alibaba, qwen.ai) — Alibaba’s family. Large and capable; deeply integrated into Alibaba Cloud.
  • Doubao (ByteDance) — chat (called Cici outside China). ByteDance also owns TikTok and CapCut.
  • Kimi (Moonshot AI, kimi.com) — long-context chat that gained attention in 2024.
  • Ernie / Wenxin (Baidu) — China’s earliest serious LLM family.
  • MiniMax (hailuoai.com chat side) — chat + video gen.
  • Zhipu GLM / ChatGLM (zhipuai.cn) — chat from Zhipu AI.
  • Tencent Hunyuan — Tencent’s family; embedded in WeChat.
  • 01.AI / Yi (01.ai) — founded by Kai-Fu Lee.
  • Stepfun (Step) (stepfun.com) — multimodal models.
  • Manus — agent product; surged in early 2025 then faced credibility concerns.

Video / image / music / agent products

  • Kling AI (Kuaishou) — video gen; competes with Sora, Veo, Runway.
  • Hailuo (MiniMax) — video gen.
  • CapCut AI (ByteDance) — video editing app with AI features. Especially worth avoiding for business work — uploads route through ByteDance.
  • TikTok Symphony — AI ads / creative tool from ByteDance.
  • WeChat AI features (Tencent) — embedded in WeChat the messenger.

Cloud platforms

  • Alibaba Cloud (Aliyun) — China’s biggest cloud. AI services include Tongyi.
  • Tencent Cloud — AI services include Hunyuan.
  • Baidu AI Cloud — AI services include Ernie.
  • Huawei Cloud / Pangu — Huawei’s cloud + AI models.

Use AWS, Azure, GCP, Oracle Cloud, IBM Cloud, or Cloudflare instead.

Hardware

Chinese AI chips — Huawei Ascend, Cambricon, Biren, Moore Threads — are not generally available outside China and are subject to US export controls in either direction. Stick with Nvidia, AMD, Apple Silicon, Google TPU, Groq LPU, Cerebras.

What about open-weight Chinese models?

This is the genuinely subtle case. Some Chinese AI labs — DeepSeek, Alibaba (Qwen), Zhipu (GLM), 01.AI (Yi), Tencent (Hunyuan-Large) — release the actual model weights (the underlying neural-network files) for free download. Anyone in the world can take those weights, run them on their own laptop using software like Ollama 🟥 or LM Studio 🟥, or host them on a Western cloud like Together AI, Fireworks AI, Hugging Face Inference, or AWS Bedrock.

If you run the weights yourself — locally or on a Western host — the data does NOT go to China. That removes reason 1 from the list above.

But the other reasons still apply:

  • The model was trained in China under CAC content-filtering rules. The political filtering is baked into the weights themselves — running it locally doesn’t remove it.
  • Open-weight Chinese models have been shown to behave subtly differently in code generation on sensitive topics (e.g. quietly inserting bad cryptography in some niche cases), though this is an active research area.
  • Trust in the training data is unverifiable — you cannot inspect what 10+ trillion tokens the model was trained on, and the labs do not publish full training-data lists.

The encyclopedia’s recommendation: even for open-weight models, prefer Western alternatives. Meta’s 5 🟥, Mistral models 🟥, Google’s Gemma 🟥, and OpenAI’s open-weights “gpt-oss” releases 🟥 cover essentially every use case where someone might reach for a Chinese open-weight model, without the political-filtering and training-trust concerns.

The open-weights-vs-closed.md entry has the longer comparison.

”But I really need to look at one to understand what it does”

If you genuinely need to evaluate a Chinese AI tool (you’re writing a comparison article, doing security research, evaluating competitive products) and not just casually using it, sensible sandboxing:

  1. Never sign up with your real email. Use a disposable address (SimpleLogin, Fastmail aliases, addy.io).
  2. Never paste real work, code, customer data, or anything sensitive. Use synthetic test prompts.
  3. Don’t install mobile apps on your personal phone. Use a separate device or an Android emulator on a Mac/PC.
  4. Don’t pay with your real card. If a card is required, use a single-use virtual card (Wise, Revolut, ANZ Plus virtual card, Up Bank virtual card).
  5. Treat the session as 100% public. Assume Chinese state agencies will see everything you type.
  6. Don’t grant browser permissions (camera, microphone, location, clipboard).

This is the same hygiene any security researcher would apply to studying any potentially-hostile software.

Why this isn’t anti-Chinese — it’s pro-rule-of-law

The encyclopedia distinguishes between:

  • Chinese people, who are not the target of this recommendation in any way — many are deeply opposed to the same surveillance regime they live under.
  • Chinese researchers and ideas, which are essential — much of modern AI research comes from Chinese-origin researchers (many of whom now work at OpenAI, Anthropic, Google, Meta, in the US and Europe), and their published papers and open-source contributions are gold.
  • Chinese companies operating in mainland China under PRC law, which is what this recommendation is about — the legal and regulatory environment those companies operate in does not permit Western-style privacy guarantees, no matter how well-intentioned the engineers building the product.

If a Chinese-founded AI company relocates to the US or EU and is no longer subject to PRC data-handover law, the recommendation against it should be reconsidered. (None of the major Chinese AI labs has done this as of mid-2026.)

The same standard applies to any country whose legal regime forces tech companies to assist state surveillance without recourse — it’s not a “China is special” rule; it’s a “rule of law matters” rule.

Common gotchas

  • The English UI fools people. DeepSeek’s, Qwen’s, Doubao’s apps all have polished English UIs. The UI language has nothing to do with where servers are or who operates them. Look at the company headquarters, not the screen.
  • App-store availability fools people too. A Chinese app being on the Apple App Store or Google Play does not mean Apple or Google has vetted it for data-residency. They’ve checked it for malware and TOS compliance, nothing more.
  • “Hosted on AWS” doesn’t help if the company is Chinese. If DeepSeek runs an API on AWS Tokyo, Chinese law still applies to DeepSeek the company — it can still be compelled to hand over data. Use the model via a Western reseller (e.g. Together AI hosting the open-weight DeepSeek), not via DeepSeek directly.
  • TikTok is in the same legal category as Doubao — both are ByteDance. The CapCut, TikTok, and Doubao apps share infrastructure and ByteDance corporate data-handling.
  • Open-source ≠ safe. “Open-weights” means you can download the weights. It does not mean the training data was clean, the political filtering was removed, or the model is harmless.
  • VPNs do not change the legal situation. A VPN hides your traffic from your local ISP; it does not change what the receiving company is legally required to do with your data.
  • AUS government departments and many AUS large enterprises already ban these tools — DeepSeek was banned from Australian government devices on 4 February 2025 (Department of Home Affairs direction). Many AUS universities, banks, and ASX-listed companies followed suit.

See also

Sources

Graph View

Backlinks