Does ChatGPT train on your data? The training switches, temporary chats and retention periods for ChatGPT, Claude, Gemini, Copilot and more

Disclosure: NOXI publishes this guide and is one of the tools it mentions; the settings and the method work whichever assistant or presentation software you use.

Three switches decide what happens to anything you paste into an AI assistant: whether training is on, whether you are in a temporary or incognito chat, and whether the account is personal or business. This guide gives the exact name and path of each switch for ChatGPT, Claude and Gemini, then Copilot, Perplexity, Mistral and DeepSeek, how long each keeps your chats, what business plans change, and a worked example in which the same paste lasts 3 days or 5 years. Every setting name, default and retention period comes from the vendor's own help or privacy page, opened on 22 October 2026.

Key numbers

All six come from the vendors' own pages; the sections below show which switch moves each one.

How AI assistants train on your chats, in plain words

When an assistant "trains on your data", the vendor adds your conversations to the material it uses to improve future versions of its models. Once used, it leaves your account's control: vendors say data already used in training cannot be pulled back out of a model that has been trained on it.

Anthropic states the limit plainly: after you turn its setting off, your data stays in training runs already in progress and in models already trained, though stored chats are no longer used in future runs [8]. Perplexity says previously collected training data cannot be deleted or removed, and its opt-out applies only to data collected after the opt-out date [19]. So the setting matters most before you paste, not after.

Three things that can send a chat to training even with the setting off

1. Feedback buttons. On OpenAI's individual plans, a thumbs up or down means "the entire conversation associated with that feedback may be used to train" its models, even if you opted out [2]. At Mistral, a rating with a comment lets it use that rating, input and output, on any plan [20]. Anthropic keeps feedback data for 5 years [7].
2. Safety review. Anthropic may use chats flagged by its safety classifiers to improve its trust and safety systems, and keeps flagged inputs and outputs up to 2 years and classification scores up to 7 years [7][8].
3. Human review. Google's human reviewers see some Gemini chats; reviewed chats are disconnected from your account, kept up to 3 years and not deleted when you delete your activity [13].

Google's own Privacy Hub puts the practical rule in one line: "Please don't enter confidential information that you wouldn't want a reviewer to see …" [13].

The settings table: training defaults, switches and temporary chats in 7 assistants

On personal plans, ChatGPT, Gemini, Perplexity and Mistral use chats for training by default until you switch it off; Claude trains only if Model Improvement is on; the new Microsoft Copilot app says prompts, responses and files are not used to train foundation models. Every business offer in the last column excludes business content from training by default; DeepSeek's consumer policy describes none.

Read the table by row before you paste. "Not documented" means the official page we opened on 22 October 2026 does not describe such a mode; it is a statement about the page, not a finding that the feature is absent.

Does ChatGPT train on my data? How to turn it off

Yes, ChatGPT's personal plans may train on your conversations by default. OpenAI says that on its services for individuals "we may use your content to train our models", and you opt out by turning off Improve the model for everyone in Settings → Data controls, or by selecting Do not train on my content in the OpenAI Privacy Portal; either one is enough [2].

OpenAI's help article on ChatGPT Go repeats the same line for Go: conversations may be used to improve model performance, and you can opt out at any time [5]. Plan does not change the default for individuals: Free, Go, Plus and Pro all start with training allowed.

How to turn off ChatGPT training on the web, iPhone and Android

The setting follows your account across web and mobile once you are signed in. These are the steps in OpenAI's article "Data controls in ChatGPT", updated on 22 October 2026 [1].

Three details from the same article are easy to miss. Turning the setting off applies to new conversations; it does not delete or hide saved chats. Archiving a chat does not change whether it can be used for training. And memory is a separate setting: turning memory off does not turn training off, and the other way round [1].

ChatGPT temporary chat: what it keeps and for how long

A ChatGPT temporary chat is not used to train OpenAI's models while it stays temporary, does not appear in your history and does not create or update memories; OpenAI may keep a copy for up to 30 days for safety purposes [3]. Start one by opening a new chat, selecting Temporary and choosing Personalized or Unpersonalized before your first message [3].

One action undoes it: if you save a temporary chat, it becomes a regular chat and follows your account's model-improvement setting [3]. If you use a GPT with actions in a temporary chat, data sent to that third party follows the recipient's own privacy policy and may be kept longer than 30 days [3].

What happens when you delete a ChatGPT chat

Deleting a saved chat removes it from your view at once, and OpenAI schedules it for permanent deletion from its systems within 30 days, unless it was already de-identified and disassociated from your account or OpenAI must keep it longer for security or legal reasons [4]. A file saved to Library is not deleted with the chat; delete it from Library separately [4].

Does Claude train on my data? Model Improvement, the 2025 change and incognito chats

Claude trains on your Free, Pro or Max chats only if you allow it through the Model Improvement setting; Anthropic's privacy centre frames training as something you "choose to allow" [6]. If you allow it, de-identified chats can stay in Anthropic's training pipelines for up to 5 years; if you do not, deleted chats leave its back-end storage within 30 days [7].

The 2025 consumer terms change

On 28 August 2025 Anthropic updated its consumer terms and privacy policy for Claude Free, Pro and Max, including Claude Code used from those accounts, and asked existing users to choose a model training setting by 8 October 2025 to keep using Claude [11]. Choosing to allow training extended retention to five years; not allowing it kept the existing 30-day period [11]. The change applied only to new or resumed chats and coding sessions, and it did not apply to Claude for Work (Team and Enterprise), Claude for Government, Claude for Education, the API, or access through Amazon Bedrock and Google Cloud's Vertex AI [11].

How to turn off Claude training

On the web or Claude Desktop: open the menu under your name → Settings → Privacy → under Help Improve our AI models, switch the toggle off. On a phone: your name → Settings → Privacy → the same toggle [8].

After you switch it off, new chats and coding sessions are not used for future training, and stored chats stop being used in future training runs; data already in a run that has started, or in a model already trained, stays there [8]. Deleting a chat also removes it from future training [7].

Claude incognito chat

A Claude incognito chat is not saved to your history or to Claude's memory and is not used for training, even if Model Improvement is on [7][10]. It is kept for 30 days by default, or longer under an Enterprise organisation's custom retention setting [10]. Start one outside a project by clicking the ghost icon in the upper right corner of a new chat; incognito chats are available on Free, Pro, Max, Team and Enterprise [10].

Claude data retention: 30 days, 2 years, 5 years, 7 years

Anthropic's consumer retention article, dated 1 July 2026, lists four periods [7]:

• 30 days: deleted chats are removed from back-end storage within 30 days.
• 5 years: de-identified chats in training pipelines when Model Improvement is on; feedback you submit with the thumbs buttons or a bug report.
• 2 years: inputs and outputs of chats flagged as violating the Usage Policy.
• 7 years: trust and safety classification scores for flagged chats.

Is Claude safe for sensitive data?

The same three checks apply: Model Improvement off, an incognito chat, and a Team or Enterprise account for work data, whose content is not used for training by default [9][10]. What you paste is still kept on Anthropic's systems for the periods listed above.

Does Gemini train on my data? Keep Activity, auto-delete and temporary chats

Yes, while Keep Activity is on. Google's Gemini Apps Privacy Hub, last updated on 19 October 2026, says Google uses your activity to provide, develop and improve its services, "including training generative AI models", with help from human reviewers [13]. Keep Activity is on by default if you are 18 or over [14].

How to turn off Gemini Apps Activity

On a computer, go to gemini.google.com → Settings & help → Activity → near the top, click On → Turn off, or Turn off and delete activity; the same page is at myactivity.google.com/product/gemini [14]. With Keep Activity off, future chats do not appear in your activity and are not used to train Google's AI models unless you send feedback, but they are still saved for 72 hours so Gemini can respond, process feedback and protect users [13]. Some Connected Apps are unavailable while the setting is off [13].

Gemini auto-delete and what deleting does not remove

With Keep Activity on, activity is auto-deleted after 18 months by default; you can choose 3 or 36 months, or not to auto-delete [13][14]. Deleting activity removes it from your account, but chats already reviewed by human reviewers are not deleted, because they are no longer connected to your account; they are kept for up to 3 years [13]. Turning off Keep Activity also does not delete data saved by other Google services, such as Search Services History [13].

Gemini temporary chats

Gemini temporary chats are kept with your account for 72 hours and are not used to train Google's AI models [13]. If you send feedback while Keep Activity is off, Google collects the feedback and the last 24 hours of your chats as context [13].

A work or school Google Account follows different terms, set by your Google Workspace administrator [13][14]; the business section below covers them.

Copilot, Perplexity, Mistral and DeepSeek: does each train on your data?

Microsoft's new Copilot app says it does not train foundation models on your chats; Perplexity and Mistral train by default with an opt-out toggle, and Mistral adds a Temporary Chat; DeepSeek's privacy policy lists training as a use and states an opt-out right without naming a setting.

Does Microsoft Copilot train on my data?

For the updated Microsoft Copilot app on web, desktop and mobile, available from 18 August 2026, Microsoft says prompts, responses and your file contents aren't used to train foundation models; optional feedback may improve Copilot but is not used to train them either [16]. This applies when you sign in with a personal Microsoft account. If you have not updated the app, Microsoft's earlier privacy FAQ applies: it offers an opt-out from model training and stores conversation activity for 18 months by default [17].

Does Perplexity train on my data?

Yes, by default on Free, Pro and Max: Perplexity's help centre says AI Data Retention is enabled by default [19]. Turn it off in Account settings → Preferences → the AI data retention toggle. The opt-out covers only data collected afterwards, and the setting applies to individual users [19]. Perplexity Enterprise data is never used for AI training, and Enterprise uploads are kept for 7 days [19].

Does Mistral's Vibe train on my data?

Mistral's assistant, now called Vibe, uses input and output data to train its models by default unless you opt out [20]. On the web, open the Admin panel → Manage → Vibe → Privacy and switch off "Allow your interactions to be used to train our models"; on iOS and Android, open Settings → Data & Account Controls and deselect Enable data sharing [21]. Mistral notes that uploaded documents count as input data, and that the Vibe and API opt-outs are separate toggles [21].

Is DeepSeek safe to use?

That depends on what you paste: DeepSeek's privacy policy, last updated 10 February 2026, says it collects, processes and stores your personal data in the People's Republic of China and uses data to train and improve its models [22]. It lists a right to opt out of training, by email to privacy@deepseek.com, but names no in-app switch; you can copy or delete your chat history in settings [22]. Treat it as an account with training on.

Grok and Meta AI sit outside the table. Grok's consumer FAQ names an "Improve the Model" setting and a Private Chat whose content is not used for model training [25]; Meta says AI interactions will be used for training, with an objection form for people in the EU [26].

Does ChatGPT Business or Enterprise train on your data? What business plans change

No. OpenAI says that by default it does not use content from ChatGPT Business, Enterprise, Edu or ChatGPT for Healthcare workspaces to train its models, and Anthropic, Google and Microsoft say the same of their business offers, as do Perplexity and Mistral of their Enterprise plans [1][9][15][18][19][21].

A business plan changes who decides. In ChatGPT Business and Enterprise, data export and account deletion go through the workspace owner, not self-service [1]; in Claude Team and Enterprise, members can export their own chats only once an owner turns that on [12]. Your organisation's retention and access policies apply on top of the vendor's [1].

Is ChatGPT safe for confidential information? A decision list

A personal ChatGPT account with training on is the wrong place for client, patient or employer data; a business workspace your employer has approved, or at minimum a personal account with training off and a temporary chat, keeps the content out of training. Neither choice decides whether you are allowed to share the data at all: your contract, your employer's policy and the law where you work decide that.

This is not legal advice. It is a sequence of checks built from the vendors' stated settings.

Eight passes mean the vendor's own settings are working for you. One fail on items 1, 2 or 8 means stop and ask, whatever the other switches say.

ChatGPT vs Claude privacy, in one paragraph

On personal plans the main difference is the starting point: ChatGPT may train until you turn Improve the model for everyone off [2], while Claude trains only if Model Improvement is on [6]. Both offer a no-training chat mode kept about a month (ChatGPT temporary chat up to 30 days, Claude incognito 30 days by default) [3][10], both remove deleted chats within 30 days [4][7], and both exclude business plans from training by default [2][9]. The longest period on either side is Anthropic's 5 years for training data and feedback [7]. Our ChatGPT vs Claude comparison scores the rest.

Is it safe to upload documents to ChatGPT? Work, tax and legal files

An uploaded file is treated like the rest of the conversation: Mistral states that attached documents count as input data [21], Google lists uploads as part of the activity Keep Activity saves [13], and OpenAI's training setting covers the content of your conversations [2]. If training is on, assume the file is in scope.

• Tax returns and payslips: remove tax IDs, bank details and addresses first, or type the figures you need into the chat instead of uploading the document.
• Legal documents: contracts, court papers and advice from your lawyer can carry confidentiality duties; use an approved business account or none.
• Work files: follow your employer's policy; if there is none, ask before uploading anything a customer, colleague or regulator would not expect to see outside the company.
• Medical records: keep your own and other people's health information out of personal accounts.

Check where the file lives after the chat: ChatGPT can keep uploaded or created files in Library separately from the conversation, so deleting the chat does not delete them [4]. For preparing confidential material for slides specifically, our guide to using confidential material with AI presentation tools covers redaction and labelling step by step.

How long each assistant keeps your chats: one paste, seven outcomes

The same paste can be gone in 3 days or kept for 5 years. The difference comes from the training switch, the chat mode and whether a human reviewed the chat, more than from the assistant you picked.

The arithmetic: 18 months from 1 December 2026 is 1 June 2028, which is 365 + 183 = 548 days. Three years to 1 December 2029 is 365 + 366 + 365 = 1,096 days, because 2028 is a leap year. Five years to 1 December 2031 is 1,096 + 365 + 365 = 1,826 days. The longest outcome is 1,826 ÷ 3 = 608.7, so about 609 times the shortest.

Two switches explain most of the spread: Gemini's Keep Activity moves the same chat from 3 days to 548, and Claude's Model Improvement moves the last copy of a deleted chat from 30 days to 1,826. The third factor, human review, is not a switch you control, which is why Google advises keeping confidential information out [13].

What Calder & Moss found when it checked

Before the staff briefing, Mei asked all 64 staff two questions. 24 of 64 said they used a personal assistant account for work tasks: 24 ÷ 64 = 37.5%. Of those 24, 15 had never changed the training setting, so training was on: 15 ÷ 24 = 62.5%; the other 9, or 37.5%, had turned it off. The firm's rule after the briefing: client documents only in the firm's approved business workspace, never in a personal account; for anything else in a personal account, training off and a temporary or incognito chat.

Here are two slides from this article's worked example, as NOXI built them, unedited.

How to export or delete your data in ChatGPT, Claude, Gemini and Copilot

Each assistant lets individual users export or delete their data from settings: ChatGPT for Free, Go, Plus and Pro, Claude for Free, Pro and Max on the web or desktop, Gemini through Gemini Apps Activity, and Copilot through Microsoft's privacy dashboard.

Deleting removes the copy in your account; it does not reach into a model already trained on your data [8], so turning training off first matters more than deleting later.

For presentations: what NOXI's privacy policy says

NOXI's privacy policy, last updated on 20 September 2026, states: "We do not use your presentations, prompts, or uploaded files to train or fine-tune AI models." [23] Data is encrypted in transit and at rest, and NOXI is operated from Abu Dhabi, United Arab Emirates [23].

The policy describes one related use: if you rate a slide, a layout-only record with none of your text in it may, once reviewed and approved, become a design example containing placeholder text only [23]. The same eight checks above still apply to a deck: decide what the file may contain before you upload it.

For choosing a tool for decks rather than for chat, see our comparisons of a dedicated deck maker with ChatGPT for presentations and with Claude for presentations.

3 AI prompts for an AI privacy briefing deck

Use them in NOXI after attaching your own policy document, or in any AI tool that accepts files. Replace the brackets; the tool should arrange your facts and add none of its own.

How to build an AI privacy briefing deck in NOXI from your own policy

NOXI turns your own document into a structured deck you can edit by chat. It is free, with no card and no watermark; uploading files needs a free account, and a full deck uses about 25 of the 1,000 monthly credits. This flow uses one Home action, Create from file.

For the briefing itself, our training presentation guide covers timing and the questions slot, and the Word to PowerPoint guide and PDF to presentation guide explain how a policy document becomes slides. For a partners' meeting, the board meeting presentation guide shows how to ask for a decision on the rule.

Frequently asked questions

Does ChatGPT train on my data?

Yes, on personal plans by default. OpenAI says it may use content from its services for individuals, including Free, Go, Plus and Pro, to train its models. Turn off Improve the model for everyone in Settings, Data controls, to stop new chats being used. Business, Enterprise and Edu workspaces are not used for training by default.

How do I stop ChatGPT from training on my chats?

Open Settings, select Data controls and turn off Improve the model for everyone, or select Do not train on my content in OpenAI's Privacy Portal; either is enough. The choice covers new chats on every signed-in device. Avoid thumbs-up or thumbs-down feedback, because rating a response can send that whole conversation to training.

Does Claude train on my data?

Only if you allow it. On Claude Free, Pro and Max, Anthropic trains on chats only when the Model Improvement setting is on, found under Settings, Privacy, Help Improve our AI models. Incognito chats are never used for training, and Team, Enterprise and API content is not used for training by default.

How long does Claude keep my data?

Deleted Claude chats leave Anthropic's back-end storage within 30 days. With Model Improvement on, de-identified chats can stay in training pipelines for up to 5 years, and feedback is kept 5 years. Chats flagged for Usage Policy violations are kept up to 2 years, with safety classification scores kept up to 7 years.

Does Gemini use my chats for training?

Yes, while Keep Activity is on, which is the default for users aged 18 and over. Google says it uses that activity to improve its services, including training generative AI models, with help from human reviewers. Temporary chats and chats made with Keep Activity off are not used to train Google's models.

What happens when I turn off Gemini Apps Activity?

Your future chats stop appearing in your activity and are not used to train Google's AI models unless you send feedback. Google still keeps them with your account for 72 hours to respond and protect users. Chats that human reviewers already saw are kept up to 3 years, even if you delete your activity.

Is ChatGPT safe for confidential information?

A personal account with training on is not the place for client, patient or employer data. Use your employer's approved business workspace, or at least turn training off and use a temporary chat, kept up to 30 days. Whether you may share the data at all depends on your contract, policy and local law.

Does ChatGPT Business train on our data?

No, not by default. OpenAI says it does not use content from ChatGPT Business, Enterprise, Edu or ChatGPT for Healthcare workspaces, or from its API, to train its models. Thumbs feedback in those workspaces does not permit training on the conversation, and your organisation's own retention and access policies also apply.

Is DeepSeek safe to use?

Judge it by what you paste. DeepSeek's privacy policy, updated 10 February 2026, says it stores personal data in the People's Republic of China and uses data to train its models. It states a right to opt out of training by email but names no in-app setting, so treat chats there as training-on.

Does NOXI use my presentations to train AI?

No. NOXI's privacy policy, last updated 20 September 2026, says it does not use your presentations, prompts or uploaded files to train or fine-tune AI models. Data is encrypted in transit and at rest, and the service is operated from Abu Dhabi, United Arab Emirates.

Sources & methodology

How this guide was made: every setting name, menu path, default and retention period comes from the vendor's own help centre, privacy centre or privacy policy, opened on 22 October 2026, several of them in a browser because they render by script. "Updated" dates are the ones each page showed that day. Calder & Moss LLP, Mei Tan and the staff survey figures are fictional. This guide explains settings and is not legal advice.

• [1] Improve the model for everyone (web, iOS, Android, signed out); Do not train on my content in the Privacy Portal; temporary chats not used to train, kept up to 30 days; saved chats unaffected; feedback after opt-out; export and deletion for Free, Go, Plus and Pro; managed workspaces not used to train by default; page updated 22 October 2026 — OpenAI Help Center, Data controls in ChatGPT.
• [2] Services for individuals (Free, Plus, Pro) may be used to train; opt-out options; feedback trains on the whole conversation; Business, Enterprise, Edu and API not used by default; business feedback does not permit training; page updated 20 October 2026 — OpenAI Help Center, How your data is used to improve model performance.
• [3] How to start a temporary chat; Personalized and Unpersonalized; not used to improve models while temporary; copy kept up to 30 days; saving converts it to a regular chat; GPT actions; page updated 20 October 2026 — OpenAI Help Center, Temporary chat in ChatGPT.
• [4] Deleted chats scheduled for permanent deletion within 30 days, with exceptions; Library files kept separately; page updated 19 October 2026 — OpenAI Help Center, Chat and file retention in ChatGPT.
• [5] Data privacy line for ChatGPT Go (conversations may be used to improve model performance; opt out at any time); page updated 22 October 2026 — OpenAI Help Center, What is ChatGPT Go?.
• [6] Consumer products (Free, Pro, Max); training when you choose to allow it; Model Improvement; incognito chats not used even with it on; feedback; dated 16 March 2026 — Anthropic Privacy Center, Is my data used for model training? (consumer).
• [7] Deleted chats removed from back-end within 30 days; de-identified training data up to 5 years; incognito chats; flagged chats 2 years and scores 7 years; feedback 5 years; dated 1 July 2026 — Anthropic Privacy Center, How long do you store my data?.
• [8] Settings → Privacy → Help Improve our AI models on desktop, browser and mobile; what turning it off changes; flagged chats; dated 3 August 2026 — Anthropic Privacy Center, How do I change my model improvement privacy settings?.
• [9] Commercial products: inputs and outputs not used for training by default unless you report feedback or bugs or otherwise allow it; feedback kept up to 5 years; dated 18 August 2026 — Anthropic Privacy Center, Is my data used for model training? (commercial).
• [10] Incognito chats not saved to history or memory, not used for training, kept 30 days by default or per Enterprise custom retention; available on all plans; ghost icon — Claude Help Center, Use incognito chats.
• [11] Consumer terms update rolled out 28 August 2025 for Free, Pro and Max; choice required by 8 October 2025; five-year vs 30-day retention; new or resumed chats only; Claude for Work, Gov, Education, API, Bedrock and Vertex AI not affected — Anthropic, Updates to Consumer Terms and Privacy Policy (28 August 2025).
• [12] Export from Settings → Privacy → Export data on web or desktop, not iOS or Android; link expires after 24 hours; Team and Enterprise member exports need an owner setting — Claude Help Center, Export your Claude data.
• [13] Activity used to improve services including training generative AI models; human reviewers; reviewed chats kept up to 3 years and not deleted with activity; temporary chats and Keep Activity off kept 72 hours and not used to train; auto-delete default 18 months; work and school accounts; last updated 19 October 2026 — Google, Gemini Apps Privacy Hub.
• [14] Keep Activity on by default for users 18 and over; Settings & help → Activity → Turn off; auto-delete at 3, 18 or 36 months or never; deleting by item, day or range; myactivity.google.com/product/gemini — Google, Manage & delete your activity in Gemini Apps.
• [15] Content not human reviewed or used for generative AI model training outside your domain without permission; last updated 6 October 2026 — Google Workspace Help, Generative AI in Google Workspace Privacy Hub.
• [16] Updated Copilot app available from 18 August 2026; prompts, responses and file contents not used to train foundation models; feedback; export and delete history in the privacy dashboard; personal Microsoft accounts — Microsoft Support, Microsoft Copilot for individuals: your activity history.
• [17] Previous Copilot app: opt-out of model training; conversation activity stored for 18 months by default — Microsoft Support, Privacy FAQ for Microsoft Copilot.
• [18] Work or school accounts: prompts and responses not used to train foundation models; logged and searchable by IT admins — Microsoft Support, Data protection when using Microsoft Copilot Chat for work or school.
• [19] AI Data Retention on by default for Free, Pro and Max; Account settings → Preferences toggle; opt-out applies only to later data; previously collected training data cannot be removed; Enterprise data never used for training; Enterprise uploads kept 7 days; last modified 31 August 2026 — Perplexity Help Center, Data Collection at Perplexity.
• [20] Vibe: input and output used to train by default unless you opt out; Vibe Enterprise opted out by default; a rating with a comment authorises use of that input and output on any plan; page updated 21 October 2026 — Mistral Help Center, Do you use my user data to train your Artificial Intelligence models?.
• [21] Opt-out paths on web (Allow your interactions to be used to train our models) and mobile (Enable data sharing); Teams admins; Enterprise default; documents count as input; separate Vibe and API toggles; dated 3 September 2026 — Mistral Help Center, Can I opt out of my input or output data being used for training?.
• [22] Data used to train and improve models; storage in the People's Republic of China; right to opt out of training; copy or delete chat history in settings; last update 10 February 2026 — DeepSeek, Privacy Policy.
• [23] "We do not use your presentations, prompts, or uploaded files to train or fine-tune AI models"; layout-only rating records; encryption in transit and at rest; operator in Abu Dhabi, UAE; last updated 20 September 2026 — NOXI, Privacy Policy.
• [24] Temporary Chat "isn't saved to your history or used to train our models" — Mistral docs, Vibe.
• [25] "Improve the Model" setting; Private Chat content not used for model training — xAI, Consumer FAQs.
• [26] Meta AI interactions used to train; objection form for people in the EU (14 April 2025, updated 27 March 2026) — Meta, Making AI work harder for Europeans.

Brief your team on AI privacy with NOXI, free: attach your AI use policy, check the switches and the retention slide in Plan mode, and export an editable PowerPoint or a PDF.