⏱️ Lectura: 11 min
Mistral automatically enables training with your conversations in Vibe, its consumer assistant: to stop it, you have to go into the privacy panel and turn off a specific switch. The training opt-out exists, but it isn’t a single button: there are two separate controls, one for Vibe and another for Mistral Studio and the API, and you need to enable them separately if you use both products.
📑 En este artículo
- TL;DR
- What Mistral says about the training opt-out
- Context: why this control exists and what the regulations say
- Technical details: how the two toggles work
- How to enable the opt-out step by step
- Impact and analysis for teams in Latin America
- What’s next
- Frequently Asked Questions
- Does Mistral use my conversations to train its models by default?
- Are the documents I upload to Vibe also used for training?
- If I disable the opt-out in Vibe, is it also disabled in the API?
- How do I disable training from my phone?
- What happens to data that was already used for training before I enabled the opt-out?
- References
The company documents this in its official help center: in Vibe (consumer version), using your interactions to train models is enabled by default, while in Vibe Enterprise the opt-out already comes applied from the start and is managed by whoever administers the account. For teams in Latin America integrating the Mistral API into their own products, understanding this difference is key before exposing end-user data.
TL;DR
- In Vibe (consumer), Mistral uses your conversations to train models by default: the opt-out is enabled manually in Settings at any time.
- In Vibe Enterprise, customers come with the training opt-out enabled by default: it’s managed by the account administrator.
- The Vibe toggle is called “Allow your interactions to be used to train our models” and is located in Manage, Vibe, Privacy within the Admin panel.
- Documents you attach or upload in Vibe count as input data and are also covered by that same toggle.
- In the mobile app (iOS and Android), the opt-out is done in Settings, Data & Account Controls, by unchecking “Enable data sharing”.
- Mistral Studio and the API have a different toggle, called “Anonymous improvement data”, located in Privacy within the Admin panel.
- The two toggles are independent: opting out of training in Vibe doesn’t automatically exclude you from training in the API, and vice versa.
What Mistral says about the training opt-out
The Mistral help center article is explicit: in certain cases, input and output data (conversations, documents, and other content provided by the user) may be included in Mistral’s model training programs. The company clarifies that users retain full control over that processing and can opt out at any time, but the exact mechanism changes depending on which product you use.
Vibe, the consumer app
For standard Vibe, Mistral confirms that users are not excluded by default: training with your data is enabled from the moment you create your account, and it’s up to you to turn it off. This contrasts with other services where explicit opt-in is the norm; here the starting point is automatic participation.
Vibe Enterprise: a different default behavior
Vibe Enterprise customers start from the opposite position: the training opt-out comes applied from day one, and it’s the organization’s administrator who decides whether to enable training for the rest of the team. This difference between the consumer plan and the enterprise plan is usually the first thing security teams ask about when evaluating Mistral for a corporate environment.
Context: why this control exists and what the regulations say
The training opt-out isn’t an isolated courtesy from Mistral: it responds to a regulatory pattern that’s already pushing the entire generative AI industry. The European Union’s General Data Protection Regulation (GDPR) requires companies to give users granular control over how their personal data is processed, including the right to object to certain uses. Mistral, headquartered in France, operates under direct GDPR jurisdiction, which explains why the opt-out is available without friction for any account, not just enterprise customers with special contracts.
For developers in Latin America consuming the Mistral API from outside the European Union, this matters for two practical reasons. First, if your product serves European users, the chain of responsibility over that data includes you as a data controller, not just Mistral as a processor. Second, even if your user base is entirely Latin American, the training opt-out standard required by GDPR ended up becoming the minimum floor offered by most serious LLM providers, so it’s worth auditing as part of any vendor evaluation.
💭 Key point: the training opt-out in Vibe and in the API/Studio are completely independent controls. Configuring one doesn’t exempt you from the other.
Technical details: how the two toggles work
The simplest way to understand the system is to think of it as two separate surfaces: one for the chat product (Vibe) and another for everything that goes through Mistral Studio or direct API calls. Each surface has its own toggle, its own label text, and its own location within the Admin panel.
| Product | Default behavior | Where to disable it | Toggle scope |
|---|---|---|---|
| Vibe (consumer) | Training enabled | Admin panel, Manage, Vibe, Privacy | Conversations and documents attached in Vibe |
| Vibe Enterprise | Opt-out already applied | Managed by the account administrator | Conversations and documents for the entire team |
| Mobile app (iOS/Android) | Training enabled | Settings, Data & Account Controls | Use of the Vibe mobile app |
| Mistral Studio / API | Depends on account configuration | Admin panel, Privacy, Anonymous improvement data | API calls and Studio projects |
A detail that’s frequently overlooked: documents you upload or attach within Vibe count as input data just like the text you type. If your workflow uploads PDFs, spreadsheets, or screenshots to Vibe to summarize or analyze them, those files fall under the same training opt-out umbrella, so it’s worth reviewing before uploading sensitive documentation.
For teams integrating the model via API, the normal API call doesn’t change based on the toggle’s state: you’re still hitting the same endpoint with the same payload. What changes is whether Mistral uses those calls (and their responses) as part of its model improvement programs.
curl https://api.mistral.ai/v1/chat/completions \
-H "Authorization: Bearer $MISTRAL_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "mistral-large-latest",
"messages": [{"role": "user", "content": "Summarize this internal report in three bullet points"}]
}'This call is identical whether the training opt-out is enabled or not: the toggle doesn’t add parameters to the request, it acts at the account level in the Admin panel. That’s why, if your organization handles sensitive customer data, the audit step only needs to happen once in the account configuration, not on every individual call.
flowchart TD
A["You use Mistral"] --> B{"Which product do you use?"}
B -->|"Consumer Vibe"| C["Admin panel: Manage, Vibe, Privacy"]
B -->|"Vibe Enterprise"| D["Opt-out already enabled by default"]
B -->|"Studio or API"| E["Admin panel: Privacy"]
C --> F["Turn off: Allow your interactions to be used to train our models"]
E --> G["Turn off: Anonymous improvement data"]How to enable the opt-out step by step
The exact steps change depending on which product you use. Here are the three flows documented by Mistral.
From the Vibe Admin panel (web)
- Go into the Admin panel and select Vibe within the Manage section.
- Go to the Privacy section.
- Turn off the toggle labeled “Allow your interactions to be used to train our models”.
From the mobile app (iOS and Android)
- Open the app’s Settings screen.
- Select Data & Account Controls within the Account section.
- Uncheck the “Enable data sharing” checkbox.
From Mistral Studio and the API
- In the Admin panel, open the Privacy menu in the left sidebar.
- Find the “Anonymous improvement data” section.
- Turn off the toggle so API calls stop being used to improve the models.
⚠️ Heads up: Mistral doesn’t currently expose an API endpoint to read the status of these toggles. The only way to verify they’re turned off is to reopen the Admin panel and confirm visually.
As an internal checklist, many platform teams document the expected state of each toggle in a reference file (this isn’t an official Mistral API, it’s just for internal auditing):
# privacy-checklist.yml (internal team document, not official Mistral)
vibe_admin_panel:
location: "Manage > Vibe > Privacy"
toggle: "Allow your interactions to be used to train our models"
expected_state: false
mobile_app:
location: "Settings > Data & Account Controls"
toggle: "Enable data sharing"
expected_state: false
studio_api_admin_panel:
location: "Admin panel > Privacy"
toggle: "Anonymous improvement data"
expected_state: falseThis serves as evidence during a security audit or a compliance process with clients who require certifying that their data doesn’t train third-party models.
Impact and analysis for teams in Latin America
For startups and product teams in the region building on the Mistral API, having a free and accessible training opt-out, with no need for an Enterprise contract, lowers the barrier to adopting the model in products that handle customer data. There’s no need to negotiate a special agreement to stop feeding training with your end users’ conversations: it’s enough to turn off the corresponding toggle in the Admin panel.
The real friction point is the separation between Vibe and the API. A team that uses Vibe for internal prototyping and the API in production has to review and configure both toggles separately, something easy to overlook if only one of the two environments went through the security team. The practical recommendation is to treat the training opt-out as part of the onboarding checklist for any new AI provider, alongside reviewing the DPA (Data Processing Agreement) and the data retention policy.
What’s next
Mistral maintains related documentation on data retention, encryption in transit and at rest, and the option to enable Zero Data Retention (ZDR) for accounts that need call logs not to be stored at all. For teams that need formal certifications, the help center also references articles about SOC 2, ISO 27001, and where to find the company’s DPA. The general trend in the LLM industry points toward these privacy controls becoming increasingly granular and moving closer to the individual account level, instead of relying exclusively on negotiated enterprise contracts on a case-by-case basis.
📖 Summary on Telegram: View summary
Try it yourself: go into your Mistral account’s Admin panel right now, open Privacy, and confirm the current state of the Vibe and API toggles.
Frequently Asked Questions
Does Mistral use my conversations to train its models by default?
In Vibe (consumer plan), yes: training with your data comes enabled from the moment you create your account, and you have to disable it manually. In Vibe Enterprise, the training opt-out already comes applied by default.
Are the documents I upload to Vibe also used for training?
Yes. Mistral considers documents attached or uploaded within Vibe as input data, so they’re covered by the same toggle as text conversations.
If I disable the opt-out in Vibe, is it also disabled in the API?
No. They’re two independent toggles: the Vibe one lives in Manage, Vibe, Privacy, and the Studio/API one lives in the Privacy section of the Admin panel under “Anonymous improvement data”. You have to configure each one separately.
How do I disable training from my phone?
In the iOS or Android app, go to Settings, then Data & Account Controls, and uncheck the “Enable data sharing” checkbox.
What happens to data that was already used for training before I enabled the opt-out?
Mistral’s help article doesn’t detail a retroactive deletion mechanism: the training opt-out stops future use of your data starting from the moment you confirm it. For questions about how long data is retained, Mistral maintains a specific article on retention in its help center.
References
- Mistral Help Center: official article on how to opt out of having your input and output data used for training.
- Mistral AI: the company’s official site, with access to Vibe, Mistral Studio, and product documentation.
- Mistral AI Docs: official technical API documentation for developers.
- GDPR.eu: reference on the European Union’s General Data Protection Regulation, the regulatory framework behind this type of control.
📱 Enjoy this content? Follow @programacion on Telegram for daily tech content in Spanish: quick summaries, fresh content every day. @programacion
0 Comments