MorseSign inStart freeOpen Morse

Le Chat privacy policy history: every change Mistral made, dated

The Morse teamPublished 12 min read

Mistral has rewritten Le Chat's privacy terms at least 11 times since launching the chatbot on February 26, 2024, and as of November 2026 the same policy covers Vibe, Le Chat's new name. At launch only paying users could switch off training. On February 6, 2025, the day Le Chat Pro launched, the opt-out reached Free users; on April 8, 2026 Le Chat Team dropped off the list of plans Mistral promised not to train on; and on July 27, 2026 that promise left the policy altogether.

On November 18, 2026, we downloaded every version in Mistral's legal center (six privacy policies and seven sets of consumer terms for Europe, back to November 28, 2025), compared each with the one before it, and read Wayback Machine copies of Mistral's older legal page from February 2024 to November 2025. Mistral deserves credit for keeping a PDF of each recent version under a Versions menu, for naming its abuse-monitoring window (30 days) since day one, and for promising 30 days' notice of material changes in its consumer terms. Its 2024 legal page carried no version date at all, though, so the 2024 rows below are dated by the day the Wayback Machine saw the text. Effective dates start with the February 6, 2025 version.

When did Le Chat's privacy policy change?

The table lists each change we found that touches your chats, from Mistral's own pages. Rows marked "seen" come from a Wayback Machine copy: the text was live on that day and wasn't on the last copy we read before it.

DateWhat changed for your chatsSource
Feb 26, 2024Le Chat launches in beta. Its terms say you "authorize Mistral AI to use Your Prompts and Outputs to improve its Models," and only "if You decide to subscribe to the Paid Chat Services" can you opt out or use Incognito ModeMistral terms, archived
Seen May 29, 2024The paid opt-out is reworded. It no longer says Mistral "will not use Your Prompts and Outputs to train its Models," but that it won't use them "to get high level information (e.g. language or field of the prompts) to improve future training"Mistral terms, archived
Seen Sep 17, 2024First stated period for chat history: "for the duration of Your registration on the Website and/or until You delete such data"Mistral terms, archived
Seen Nov 16, 2024Terms split into Terms of Use plus "Additional Terms - Le Chat." Free users' data "will be de-identified and used to train or improve Our Models," and Le Chat data is kept for your registration "and for 90 days"Mistral terms, archived
Feb 6, 2025First dated policy. Free, Pro, and Student users "can opt-out of training at any time by using the feature available on your account." Team isn't trained on unless you opt in. Chats kept "in an identifying form until you delete your account"Mistral terms, archived
May 27, 2025Rewritten as tables. No training on "Le Chat Team, Le Chat Enterprise or the paid version of our APIs." A row for Memory, with "explicit consent" for health details saved as a Memory. Chats kept until you delete the conversation or your account. New export and opt-out controls in your accountMistral terms, archived
Nov 28, 2025Earliest version in today's legal center. Same terms for your chats as May 2025, with La Plateforme renamed Mistral AI StudioPrivacy policy, Nov 28, 2025
Jan 1, 2026Adds a section for US state privacy laws and a "Training Datasets" paragraph on data bought or licensed from third partiesPrivacy policy, Jan 1, 2026
Apr 2, 2026Adds a Data Protection Officer you can write toPrivacy policy, Apr 2, 2026
Apr 8, 2026"Le Chat Team" removed from the no-training list, which now reads "Le Chat Enterprise or the paid version of our APIs"Privacy policy, Apr 8, 2026
Jul 27, 2026Le Chat becomes Vibe in the text. The no-training sentence for Enterprise and paid APIs is gone, and "Input" now covers data "accessed by Mistral AI through integrations or services you have connected"Privacy policy, Jul 27, 2026
Sep 3, 2026Memory "now becoming Knowledge Base," which "will not seek to store sensitive data included in your Input." Still in forcePrivacy policy

The consumer terms changed on a separate schedule, with seven European versions since November 2025.

Could you always opt out of Le Chat training?

No. For its first months, Le Chat's own terms tied the opt-out to a subscription. The launch-day text, archived on February 26, 2024, said: "If You decide to subscribe to the Paid Chat Services, You will have the option to" turn on Incognito Mode and "Opt-out of the Mistral AI Training Data at any time." Free users got no such option in the terms, which told them to "pseudonymize Your Prompts" and leave out sensitive personal and business data instead (Mistral terms, archived). Mistral's privacy policy from the same day already gave everyone a general "right to object," with requests sent to privacy@mistral.ai (Mistral privacy policy, archived).

By November 2024 the split was written into the retention section: "If You use Our Free Services on La Plateforme or Le Chat: Your User Data will be de-identified and used to train or improve Our Models," while paid users who had "not opted-out" could switch it off "by using Your account settings" (Mistral terms, archived).

That changed on February 6, 2025, the day Mistral launched Le Chat Pro at $14.99 a month (Mistral). The privacy policy dated that day was the first with an effective date, and it said: "If You use Le Chat Free, Le Chat Pro or Le Chat Student," Mistral trains on your input, output, and feedback, but "You can opt-out of training at any time by using the feature available on your account" (Mistral privacy policy, archived). The day before, on February 5, the archived terms still had the 2024 wording (Mistral terms, archived).

Today every consumer plan has the switch. Mistral's help center, updated October 20, 2026, says that on Vibe "input and output data are used by default to train our artificial intelligence models, unless you opt out," and that Enterprise customers "are opted out of training by default" (Mistral). We list the switch's exact name on the web and in the app in which AI companies train on your data.

Since 2024, Mistral has also said an opt-out doesn't reach back. The 2024 terms said it "may only be effective for future Prompts and Outputs" (Mistral terms, archived), and the current policy says that when a request "concerns the training of our models," your rights "have technical limitations" (Privacy policy).

What changed for Le Chat Team and Enterprise?

Team lost its written exemption first, then Enterprise did. Team launched on February 6, 2025 with a promise in that day's policy: "If you use Le Chat Team: We do not use your Input, Output and Feedback to train our artificial intelligence models," unless you opt in. Every version from May 27, 2025 through April 2, 2026 said: "we do not use your Input and Output to train our artificial intelligence models when you use Le Chat Team, Le Chat Enterprise or the paid version of our APIs." The April 8, 2026 version kept the sentence but cut Team from it (Privacy policy, Apr 8, 2026). The July 27, 2026 version dropped the sentence entirely, leaving only "Your Input and Output, subject to your opt-out" (Privacy policy, Jul 27, 2026).

Mistral's help center still states the exemptions the policy dropped. Its opt-out page, updated September 3, 2026, says that on Vibe Team "Administrators can disable data training usage for the entire organization," and that Enterprise customers "are opted out of training by default" (Mistral). If you pay for a Team seat, that wording gives the switch to your admin. The policy's opening also says it "does not apply" when you use Mistral's products "in the context of your business activities," where a separate data processing agreement governs instead (Privacy policy).

How long has Mistral kept your Le Chat chats?

Until you delete the chat or your account, in every version since May 2025. Before that the rule moved around. The launch-day policy listed no period for chat history at all. It kept prompts and outputs "for 30 rolling days to monitor abuse," along with 5 years for identity data and 1 year for account data after your registration ends (Mistral terms, archived).

PeriodLe Chat historySource
Feb 2024No period stated; 30 rolling days for abuse monitoringArchived
Seen Sep 2024"For the duration of Your registration on the Website and/or until You delete such data"Archived
Seen Nov 2024Your registration "(unless You delete Your User Data from Le Chat) and for 90 days"Archived
Feb 6, 2025"In an identifying form until you delete your account"; without an account, for the length of the conversationArchived
May 27, 2025 to now"Until you delete your account or until you delete the conversation"Privacy policy

Deleting isn't instant on Mistral's side. Its help page says conversations are removed "typically after a short grace period to prevent accidental data loss, and for monitoring and legal retention purposes," and that zero data retention "is not available for Vibe Work or Chat, regardless of your subscription plan" (Mistral). The current retention list for every kind of account data is in is Le Chat private.

When did Le Chat get memory, and what did the policy say?

Mistral announced Memories on February 6, 2025, the day it launched the $14.99 Pro tier and the Team tier, as "an opt-in feature" (Mistral). It shipped to everyone on September 2, 2025, in beta and on the Free plan, with "fast import of your memories from ChatGPT" (Mistral).

The policy got there first. Its May 27, 2025 version gave Memory its own row, three months before the wide launch: details from your prompts, and health details only with "your explicit consent for any sensitive data explicitly included in your input and saved as a Memory." The September 3, 2026 policy says the feature is "now becoming Knowledge Base" and that "Knowledge Base will not seek to store sensitive data included in your Input" (Privacy policy). The product moved the other way on defaults. Mistral's documentation says "Knowledge is enabled by default," that Vibe "stores them on its own" when it spots recurring details, and that turning it off "doesn't delete your stored knowledge entries" (Mistral docs). Mistral deserves credit for one default: the setting Sensitive data in Knowledge, which lets Knowledge save details like health or religious beliefs, starts off (Mistral docs).

When did Le Chat become Vibe?

On May 28, 2026. Mistral's announcement says: "Le Chat is now Vibe," with "every conversation, setting, and plan carried over" (Mistral). Mistral's help article about the rename shows an update date of August 12, 2026 (Mistral), but four other help articles we read carry the same August 12 timestamp, to the minute. The legal text followed later: the privacy policy first says Vibe in the July 27, 2026 version, and the European consumer terms in the August 7, 2026 version.

How have Le Chat's terms of service changed?

Mistral has published seven versions for users in Europe since November 2025, with some changes in your favor and some in Mistral's. These are the versions in the legal center's EU consumer terms:

  • January 26, 2026: a "Data Switching" section lets you export "at least your Le Chat conversations (prompts and outputs)" to move to another provider (PDF).
  • March 13, 2026: Mistral "may use automated tools, including algorithmic systems," to moderate your data. Usage statistics, including your Feedback, can now serve "our business purposes, including to conduct research, improve the Mistral AI Products, create new Mistral AI Products." The line "Any such Usage Data is never used to identify you" was removed (PDF).
  • March 25, 2026: new rules for audio products, including a ban on cloning voices "without explicit consent" (PDF).
  • May 28, 2026: the Versions menu lists a consumer-terms version for this date, but its PDF opens Mistral's Additional Product Terms instead, so we can't tell you what changed (PDF as linked).
  • August 7, 2026: the training clause narrows to two cases, "(a) when you have not opted out of training or (b) when you provide Feedback to us." The November 2025 to March 2026 texts had a third, letting Mistral use data "flagged as part of our automated moderation" to improve its products. Mistral AI Studio and the APIs leave the consumer terms, now "limited to business customers" (PDF).
  • October 8, 2026: passwords, API keys, and tokens you store for agents count as your data, and you "grant us permission to use your Credentials to connect to third-party services on your behalf" (Mistral terms).

Every version we read promises at least 30 days' advance notice of changes "expected to materially impact you," by email or in-product notification. The privacy policy makes no such promise and says only "Make sure to check this Privacy Policy frequently" (Privacy policy).

The terms for users outside Europe follow a separate track, with seven versions from November 28, 2025 to October 12, 2026 (Mistral); their arbitration clause is covered in AI chatbot arbitration clauses. How Mistral's notice rules compare with other chatbots' is in AI privacy policy changes.

Has any regulator acted against Le Chat?

One has, over accuracy warnings rather than privacy. Italy's competition authority opened case PS12968 against Mistral on June 3, 2025, over whether users were told enough that its models can "hallucinate." On February 17, 2026, it accepted Mistral's commitments, which included a permanent Italian-language warning in the chat window and an Italian translation of the terms' section on hallucination risks, and closed the case "senza accertare l'infrazione," without finding an infringement (AGCM).

The current privacy policy names France's data protection authority, the CNIL, as a place you can complain and as an authority Mistral may share data with.

How Morse keeps your chats off its servers

Every row in that timeline is Mistral rewriting what it may do with chats it already holds on its servers. A new version can't take back data that was used under the old one, and Mistral's 2024 terms said an opt-out "may only be effective for future Prompts and Outputs." Morse is built so that its servers don't keep messages or log what you write, and your chat history is stored in your browser, encrypted.

  • Private, for chats you'd rather no host kept at all. It's the default for a new chat. Open-weight models like DeepSeek, Kimi, and GLM, plus Grok, run on hosts that agreed to keep nothing, because Morse has zero data retention agreements with its inference, search, and other providers. Where a provider confirms zero retention on each reply, Morse checks it and refuses a reply that doesn't.
  • E2EE, for anything you'd never want in a training set. Your browser checks the secure enclave's hardware proof before anything leaves your device, then encrypts the message so only that enclave can read it. Morse's servers relay text they can't read. It rests on the enclave hardware and attestation working as designed, and it has no web search or image generation, since models can use only tools that run on your device. E2EE chats come with Pro, at $19 a month.
  • Anonymous, for GPT, Claude, or Gemini without your account on the request. Our proxy removes your account, email, and IP address and sends the message under Morse's own account. The provider sees the message without knowing it's you, and may keep it for a limited time to check for abuse, under its own policy.

In every mode, your chat history is stored in your browser, encrypted, not on Morse's servers. Morse doesn't train AI on your chats, sell your data, or show ads (Morse privacy policy). Because we don't store chats, we usually have very little to give authorities. Clearing your browser or losing the device deletes your chats, and Morse can't recover them, so keep a copy of anything you need.

Ask a first question from the Morse home page without an account, or start free and open an E2EE chat when you move to Pro.

Common questions

Where can I read old versions of Mistral's privacy policy?

On legal.mistral.ai, under the Versions menu on each document, which keeps a PDF of every version back to November 28, 2025. Anything older, from when the terms sat on mistral.ai/terms, survives only in the Wayback Machine.

Does Mistral tell you before it changes Le Chat's terms?

For changes 'expected to materially impact you,' yes. Its European consumer terms promise at least 30 days' notice by email or in the product. The privacy policy makes no such promise and asks you to 'check this Privacy Policy frequently.'

Does opting out remove chats Mistral already trained on?

Not necessarily. The 2024 Le Chat terms said an opt-out 'may only be effective for future Prompts and Outputs,' and the current policy still warns that your rights over training 'have technical limitations.'

Is the Vibe privacy policy different from Le Chat's?

No. It's the same Mistral AI privacy policy. The July 27, 2026 version swapped the name Le Chat for Vibe throughout, two months after Mistral announced the rename.

Ask what you actually want to ask.

80+ models, characters, and images, with every chat encrypted and kept on your device. Free to start.

Start a private chat

Read next