Is Qwen safe? Who runs Qwen Chat, where your chats go, and the open weights

The Morse teamPublished October 30, 20267 min read

As of October 2026, Qwen Chat is safe for everyday questions but not for anything you need kept to yourself. Qwen's privacy policy, effective September 7, 2026, says your chats are stored in Singapore on Alibaba Cloud. The same policy says Qwen may train its models on de-identified chats. It also lets Qwen share data with its affiliates and disclose it under subpoenas, court orders, and search warrants. Qwen's open-weight models are a separate product. Anyone can download them, and once they run somewhere else, Qwen's privacy policy no longer covers the chat.

We read Qwen's privacy policy, its service agreement (effective September 10, 2026) and the February 6, 2026 version it links to, its use policy, cookies notice, and training data summary, an archived copy of the earlier privacy policy, and Qwen's model licenses on Hugging Face, all on October 23, 2026. Qwen deserves credit for naming the company that holds your data and the cloud that processes it, and for publishing most of its models under Apache 2.0.

Qwen Chat at a glance

For chat.qwen.ai, the Qwen apps, and the PC client, under the September 2026 policy and terms.

Is Qwen Chinese, and who runs Qwen Chat?

The Qwen models come from Alibaba, which Qwen's Hugging Face page calls "the large language model family built by Alibaba Cloud." The Qwen Chat app, though, is run by a Singapore company. Qwen's privacy policy names NTH POWER GLOBAL TECH SINGAPORE PTE. LTD. as the operator and data controller. The service agreement puts the contract under Singapore law, with disputes settled by arbitration in Singapore. The policy doesn't say how NTH Power relates to Alibaba. It does say your chats are processed by Alibaba Cloud (Singapore) Private Limited, and the site's footer reads "Powered by Alibaba Cloud."

Alibaba runs a different app for mainland China, called Qianwen (千问), at qianwen.com. Its privacy policy, effective October 7, 2026, names a Shanghai company as the operator. It says personal data collected in mainland China is stored in mainland China.

Does Qwen store your data in China?

Not under the current policy, but it did until September 2026. The September 7, 2026 policy says: "We collect, process, and store your personal data in Singapore." The previous version, last updated May 19, 2026 and archived on August 25, 2026, named Alibaba Cloud (Singapore) as the provider. That version said: "We store or process personal data … in Singapore and Mainland China." It added that group companies in both places had "limited remote access" to the data.

Most of the top results for "is Qwen safe" haven't caught the change. A MemX guide from August 2026 says chats are "handled under Alibaba's terms" and "can involve data being handled in China." Neither it nor the Crystal Method guide, last updated September 8, 2026, names the Singapore operator. The current policy still allows transfers to "certain affiliates and third parties" in other countries, without naming them.

Does Qwen train on your chats?

Yes, in de-identified form, and that changed in September. The February 6, 2026 terms said: "we will not use your 'Input' content provided to the Service or the 'Output' content obtained from the Service for model training purposes." The September 10, 2026 terms say Qwen may use content "to develop other products and Services, including to train models."

The privacy policy limits training to "de-identified user content and feedback data." It says Qwen "will not use your raw input data to directly train AI models." De-identifying a chat removes your name and account, but what you typed stays in the text.

There's no switch for this. The terms say you opt out by emailing privacy@qwen.com, and that feedback "will still be used for model training" afterward. When you tap thumbs up or thumbs down on a reply, the privacy policy says Qwen stores "the relevant conversations as part of the feedback." For how this compares with other apps, see which AI companies train on your data.

Who can see your Qwen chats?

Qwen staff, only in some cases. The privacy policy says Qwen won't manually review your input "unless necessary for security, fraud/abuse prevention, debugging, or compliance." It lists these other recipients:

• Affiliates, for "security, research and development, data analytics, user support, and content moderation."
• Service providers, including Alibaba Cloud (Singapore) for hosting and inference, plus log, analytics, and sign-in providers.
• Advertising partners, who get "some automatically collected device information," not chats.
• Authorities, under "subpoenas, court orders, or search warrants."
• A buyer, if NTH Power is merged, sold, or restructured.

How do you delete your Qwen data?

You delete your account yourself from Qwen's Settings page, the privacy policy says, and it gives no fixed period for keeping chats. Data is kept "only for as long as necessary," and some may be kept after you delete your account, for example under a court order or during a dispute. The Indonesia section is the only place it names a number: some data may be kept "for five years, or longer if required by law."

Is Qwen safe to run locally?

Yes, as far as Qwen's servers go. The privacy policy covers Qwen's app, website, and PC client. Weights you download send nothing to Qwen. Not every current model is Apache 2.0, though. These are the licenses Qwen's Hugging Face page showed on October 23, 2026:

None of the three limits personal use. The model Qwen Chat showed by default on October 23, 2026, Qwen3.7-Plus, has no weights on Qwen's Hugging Face page. Qwen's training data summary says the services on qwen.ai run on "proprietary models."

Running the weights yourself keeps your chats away from Qwen, but the model's training comes with them. The same summary says Qwen built "dedicated safety datasets to perform safety alignment," and whatever the model learned to refuse or slant, it still does on your machine. Our explainer on abliterated models covers what editing that out of the weights does. A 2.4-trillion-parameter model also needs server hardware, so most people end up on someone's host.

How Morse keeps your chats off the model maker's servers

Every risk above starts with Qwen Chat keeping your conversations on servers its operator controls, tied to your account, under terms that were rewritten in September. Morse's servers don't keep your messages. Morse bundles over 30 models in one account, Qwen's among them, and on Pro its image models include Qwen Image 3 Pro.

• Private, so a change in terms has no stored chat to reach. Qwen's terms went from "will not use" your input for training to an email opt-out in seven months. In Private, the default for a new chat, open-weight models run on hosts that agreed to keep nothing. Where a provider confirms zero retention on each reply, Morse checks it and refuses a reply that doesn't; otherwise it relies on the provider's commitment.
• E2EE, for drafts you'd never want in anyone's training set. Your browser checks the secure enclave's hardware attestation, then seals the message so only that enclave can read it, and Morse's servers pass along text they can't read. Third-party providers run the enclaves, not Morse, and the guarantee rests on the hardware and attestation working as designed (how the checks work).
• Anonymous, for closed models like GPT, Claude, and Gemini. Their providers see your message without knowing it's you, though they may keep it for a limited time to check for abuse, under their own policy.

Whichever mode you pick, there's no address to email about training, because Morse doesn't train AI on your chats, sell your data, or show ads (privacy policy). Your history sits in your browser, encrypted, and locks when you sign out. Because Morse doesn't store chats, it usually has very little to give authorities. Clearing your browser deletes that history for good, and Morse can't recover it, so save a copy of anything you need.

Is Qwen safe for you? A quick guide

• Where your chats are stored: Qwen Chat stores them in Singapore under a policy that changed in September. On Morse, your chat history stays in your browser.
• Training: in Qwen Chat, email privacy@qwen.com to opt out, and don't rate replies. Morse doesn't train on your chats.
• The model's built-in behavior: this goes wherever the weights go, whichever app runs them, so pick the model as carefully as the app.

Try a chat from the Morse home page without an account, or start free with no card needed. End-to-end encrypted chats come with Pro, at $19 a month.