Is it safe to use ChatGPT for translation? What each tool keeps of a client's text

The Morse teamPublished November 3, 20268 min read

It is safe to use ChatGPT for translation when the text could be published, but not for a client's confidential document on a personal plan: as of November 3, 2026, OpenAI may train on your chats on Free, Plus, and Pro unless you opt out, and its standalone ChatGPT Translate page says "Chats may be reviewed and used to improve our AI models" (ChatGPT Translate). The free versions of DeepL, Google Translate, and Microsoft Translator have their own versions of the same problem, and DeepL's free terms rule out confidential text outright.

On November 3, 2026, we read the help pages, privacy policies, and terms for ChatGPT, DeepL, Google Translate, Google's Cloud Translation API, Microsoft Translator, and Azure Translator, along with ATA's Code of Ethics, ATA's statement on AI, AUSIT's November 2025 position statement on machine translation and AI, and Article 28 of the GDPR. The paid and business versions of these tools mostly do what translators need, so the answer turns on which tier you're using. This is about privacy, not translation quality, and it isn't legal advice.

Where your source text goes, tool by tool

Every tool below gets the full text you paste or upload. The tools differ in how long they keep that text, whether they learn from it, and whether people may read it. Each cell comes from the company's own page, read November 3, 2026.

Across vendors, the free consumer tier keeps the text or learns from it, and the paid business tier usually promises not to.

Does ChatGPT train on what you translate?

Yes, on a personal plan, unless you change one setting. OpenAI's help center says that on Free, Plus, and Pro "we may use your content to train our models," and that you opt out by turning off Improve the model for everyone under Settings > Data controls (OpenAI). The opt-out covers new chats only. We cover the setting's history in does ChatGPT train on your data.

Turning training off doesn't take the text off OpenAI's servers. A saved chat stays until you delete it, and OpenAI then schedules it for deletion "within 30 days," unless it must keep it longer "for security or legal obligations" (OpenAI). A temporary chat isn't used for training, but OpenAI "may keep a copy of a temporary chat for up to 30 days for safety purposes" (OpenAI). An uploaded PDF or Word file follows its own rules, which we walk through in what happens to files you upload to ChatGPT.

OpenAI's business tiers are different. It says that "by default, we don't use inputs or outputs from ChatGPT Business, ChatGPT Enterprise, ChatGPT Edu, or our API to improve our models" (OpenAI). We compare the API with the app in is the OpenAI API private.

Is ChatGPT Translate different from a regular chat?

No. ChatGPT Translate, at chatgpt.com/translate, has its own page with a language picker, a button to "Add photos and files," and live translation, but no privacy terms of its own. The line under the box reads: "By using it, you agree to our Terms & Privacy Policy. Chats may be reviewed and used to improve our AI models" (ChatGPT Translate). The page works without logging in, and anything you paste there falls under the same rules as any other ChatGPT chat.

Is DeepL safe for confidential documents?

Only on DeepL Pro. DeepL's free terms say "you may not use the Free Services for processing content containing confidential or personal data of any kind" (DeepL terms), and its privacy policy says Pro texts are deleted once the translation is delivered and "will not be used to improve the quality of our services" (DeepL).

Does Google Translate store what you translate?

Yes, if you're signed in. Google's help center says: "When you're signed in, your translations automatically save to the cloud," and they're saved only while Web & App Activity is on. You can delete them from the app's History under Manage activity (Google).

Google's privacy policy, effective October 16, 2026, describes Translate as working "by detecting common language patterns in phrases you ask it to translate" (Google). Google's terms of service, effective July 30, 2026, give Google a license to "host, reproduce, distribute, communicate, and use your content" for purposes that include "operating and improving the services" and "developing new technologies and services" (Google). The Cloud Translation API is separate. Google says it "does not use any of your content for any purpose except to provide you with the Cloud Translation API service" (Google).

What about Microsoft Translator?

The consumer app and the business service follow different rules. Microsoft's privacy statement, last updated September 2026, says the Translator app "processes the text, image, and voice data you submit" to "improve our products and services," and that it removes identifiers such as email addresses "when randomly sampling text and audio for product improvement" (Microsoft). Azure Translator, the paid service, "doesn't persist customer data submitted for translation": text isn't stored, and documents are removed "with a hard delete" after processing (Microsoft).

What do translators' codes and NDAs say about AI tools?

They say confidentiality comes first and leave the tool choice to you. ATA's Code of Ethics, approved in November 2022, commits members "to hold in confidence, not divulge, and protect privileged and/or confidential information obtained in the course of our work" (ATA). ATA's statement on AI, approved May 3, 2025, says AI tools "may store, process, or even repurpose the data," which "can lead to breaches of confidentiality agreements and violations of data protection laws" (ATA).

Australia's AUSIT goes further. Its November 2025 position statement says translators who use AI must keep confidentiality, for example "by removing all identifying data from the source text," and it "strongly recommends" reviewing the terms of every tool that touches client text, including "email providers, cloud storage solutions, CAT tools, grammar checkers and MT/AI" (AUSIT). An NDA is narrower still. If it says the text goes to no third party, an AI company counts as one.

Does GDPR let you paste a client's document into an AI tool?

Usually not without the client's say-so, if the document holds personal data. A translator working on a client's documents generally processes that data on the client's behalf, which makes the translator a processor under Article 28, and Article 28 says "the processor shall not engage another processor without prior specific or general written authorisation of the controller" (GDPR, Article 28). An AI tool that stores the text is another processor. Birth certificates, medical records, and immigration files are typical documents that carry personal data, and we cover what AI companies keep of immigration questions in private AI for immigration questions.

A free tier whose terms let the company train on your text is hard to square with a processor's duties.

How Morse handles a client's source text

Every problem in that table starts with the source text sitting on a company's servers, under terms that let it be kept, read, or learned from. Morse is built so that, in Private and E2EE, your text isn't kept there. Your chats aren't kept on our servers, and you choose for each chat who can read it:

• E2EE, for a client's text. Your browser checks the secure enclave's hardware attestation, then encrypts the text so only that enclave can read it, and Morse's servers relay ciphertext they can't read. E2EE runs open-weight models such as GLM, Kimi, and DeepSeek. The enclaves are run by third-party providers, not by Morse, and the guarantee rests on the enclave hardware and attestation working as designed (how it works).
• Private, for terminology and phrasing questions. The default mode. Open-weight models plus Grok, on providers that agreed to keep nothing, because Morse has zero data retention agreements with its inference, search, and other providers (what that means).
• Anonymous, to use GPT, Claude, or Gemini without the provider knowing it's you. Morse's proxy removes your account, email, and IP address. The provider still sees the text and may keep it for a limited time to check for abuse, under its own policy, so keep client text out of it, or start that chat in E2EE.

In every mode, Morse doesn't train AI on your content, sell your data, or show ads, so there's no training switch to find. Our servers don't keep messages or log message content, and files you attach are sent with the message and never stored by Morse. Your history is stored in your browser, encrypted. Clearing the browser deletes it for good, and Morse can't recover it, so save each finished translation in your own files. Read our privacy policy the way AUSIT asks you to read any tool's terms.

Which tool fits which text?

• Text that's already public, such as a published web page: any tool, free tiers included.
• A client's confidential text: pasted into a Morse chat in E2EE, which rests on the enclave hardware and attestation working as designed, or a business tier your contract allows, such as DeepL Pro or Azure Translator. Skip free DeepL, whose terms forbid it, and personal ChatGPT plans.
• Personal data under GDPR: only a tool the client has authorized, as Article 28 requires.
• A question about a term or a phrase, with no client text in it: any tool, or Morse in Private mode.

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.