1. What the document is: Mistral AI's Terms of Service for Commercial Users — covers Mistral AI Studio, Le Chat, and (in later versions) Mistral Code, governing both API/Studio commercial customers and consumer-tier subscriptions (free, Le Chat Pro, Le Chat Teams) bundled under the same contract.
What this wiki found — complete, every page
Presence facts, not verdicts — each flagged term links to its definition and the exact clause on this page. Absence of a badge means the term isn't currently flagged here, not that the page is risk-free.
Le Chat – Commercial Terms
Dataset: GenGA (Generative AI Governance Archive) — 11 AI services, 2025–present
GenGA (Generative AI Governance Archive) is this wiki’s Generative-AI dataset: raw policy snapshots captured by the Open Terms Archive
genai-euproject across 11 GenAI providers (ChatGPT, Claude.ai, DeepSeek, Google Generative AI Services, Le Chat, Llama API, Meta AI, Microsoft Copilot, Perplexity, Qwen Chat, xAI). Unlike PGAv2, GenGA has no pre-tagged risk scores — all risk scoring and keyword tagging on these pages is LLM-assigned by direct reading, spanning 2025–present.
Note: This page contains documented policy clauses. Risk assessment is qualitative and context-dependent. For analysis of patterns across platforms, see:
Source: sources/GenGA/Le Chat/Commercial Terms/ (raw Markdown captures, Open Terms Archive genai-eu dataset — no pre-tagged JSONL exists for this source)
Last updated: 2026-04-08
Note on methodology: No pre-computed
risk_score/keywordfields exist for this source. Keyword tags and clause analysis on this page are LLM-assigned, using the same rubric/vocabulary as the PGAv2 pages plus new GenAI-specific tags (input license,output ownership,competing model ban,ai disclosure). Treat as first-pass analysis, not externally verified ground truth. This wiki’s risk-scoring system was retired project-wide on 2026-06-21 (seemethodology.md§4) — no page, GenGA included, computes or displays a numeric risk score.Skipped capture:
2025-12-15T19-04-35Z.md(31 bytes, body reads only “Redirecting to legal.mistral.ai”) is a failed scrape, not a real policy version — skipped and excluded from the dates below.
Overview
1. What the document is: Mistral AI’s Terms of Service for Commercial Users — covers Mistral AI Studio, Le Chat, and (in later versions) Mistral Code, governing both API/Studio commercial customers and consumer-tier subscriptions (free, Le Chat Pro, Le Chat Teams) bundled under the same contract.
2. Input rights: Reversed direction over time — the most significant finding for this provider. The 2025-11-11 baseline contained an unconditional no-training commitment (“We do not have access to the Infrastructure or Your Data and do not use Your Data to train our Models”), comparable to Anthropic’s Commercial Terms. By 2025-12-15 this had been replaced with a conditional opt-out-by-default structure with explicit training exceptions, which was then progressively broadened: 2026-03-12 added a free-subscription/Le Chat Pro training-eligible-by-default carve-out plus a Labs Models carve-out; 2026-04-08 extended the same carve-out to Le Chat Teams.
3. Output restrictions: §3.3 bans using image Outputs “to develop or train any image generation product that competes with a Mistral AI Product” — a competing model ban, but narrower in scope than OpenAI’s (image-generation only, not all Outputs). A broader, additional restriction on web-search-derived Outputs (no machine-learning use, no redistribution/resale) existed transiently in the 2025-12-15 version and was removed by 2026-03-12.
4. Non-explicit predatory clauses: §4.3 Labs Models: customers using Mistral’s experimental “labs”-prefixed models are training-eligible by default, and critically — “the opt-out preferences you selected for other Mistral AI Products does not apply to Labs Models.” A customer’s general opt-out is silently bypassed for this product tier. §14.1 (2025-12-15 version only) let Mistral create “aggregated and anonymized” Usage Data from customer use for its own business/research purposes “without constituting Customer Content” — this clause was removed entirely by 2026-03-12.
5. Regulatory references: None found — no GDPR or EU AI Act citation in this document. §14.13 (Export/Trade Controls) cites EU/US/Singapore sanctions regimes, but that is unrelated to data protection or AI regulation.
6. Regional variation: Not addressed beyond the sanctions-compliance clause.
7. Key risk to users: The directional reversal — from an unconditional no-training guarantee to a steadily-expanding set of default-training tiers and exceptions — is the central risk. The Labs Models opt-out-override is also notable as a pattern not found elsewhere in this dataset (other providers’ tier-specific exceptions don’t explicitly state that a general opt-out is bypassed).
Flagged Keywords & Risks (LLM-assigned)
input license,royalty-free— Mistral holds a perpetual, irrevocable, worldwide, royalty-free license to train on Customer Data and Outputs. The license only kicks in for certain tiers and exceptions, but those exceptions have grown broader with each revision. Why it matters: a no-training commitment eroded into a multi-exception opt-out structure within roughly four months of captures.competing model ban— image-Output competing-product training ban (§3.3). Why it matters: narrower than OpenAI’s/Anthropic’s all-Output bans, scoped specifically to image generation.feedback— Mistral can use Feedback freely and without payment, and this right explicitly extends to “the Customer Data and Output associated with such Feedback,” not just the feedback text itself. Why it matters: rating a response can pull your underlying data and Output into the same broad, unpaid usage right — wider in scope than most providers’ Feedback clauses.output ownership— Customer retains ownership of Input/Customer Data and owns all Output; Mistral assigns any rights it might have in Output to Customer. Why it matters: user-protective, consistent with most other providers in this dataset.ai disclosure— “Customer may not represent or imply that the Output was generated by a human when it was generated by the Mistral AI Products” (§3.2). Why it matters: an AI-disclosure obligation imposed on the customer (re-distributor) rather than on Mistral itself — a B2B variant of the AI-disclosure pattern seen in consumer-facing AUPs elsewhere in this dataset.
Regulatory & Research Context
Pandit et al. (2026), who directly analyzed Mistral/Le Chat’s terms, report “significant imbalance of power” in GenAI contracts generally — visible here in the §4.3 Labs Models clause, where a customer’s general training opt-out is explicitly overridden (“the opt-out preferences you selected for other Mistral AI Products does not apply to Labs Models”) without renegotiation. Applying Davidson et al.’s (2026) regulatory-gray-area framework by analogy (Mistral was not among the providers they directly studied), the directional reversal documented here — from an unconditional 2025-11-11 no-training commitment to a steadily-broadening set of default-training tiers — exemplifies the kind of post-hoc narrowing of user protections their framework treats as creating uncertainty for users who relied on the earlier commitment.
Changes Summary
| Date | What changed |
|---|---|
| 2025-11-11 | Baseline version — unconditional no-training commitment for Customer Data (“we do not use Your Data to train our Models”); Output ownership clause already present. |
| (2025-12-15 19:04, skipped) | Failed capture — page redirected to legal.mistral.ai, no content retrieved. |
| 2025-12-15 23:34 | Major substantive reversal: no-training commitment replaced by a conditional opt-out-by-default training license, with exceptions for Feedback, moderation-flagged content, and Order Form terms. New §3.3(b) restricted web-search Output use (no ML use, no redistribution). New §14.1 “Usage Data” anonymized-dataset clause. Terminology shifted from “Input”/“Customer Content” to “Customer Data”/“Outputs” throughout. |
| 2026-03-12 | Training exceptions broadened: added free-subscription/Le Chat Pro default-training tier and a new §4.3 Labs Models carve-out (with explicit opt-out override). The §3.3(b) web-search Output restriction was removed. The §14.1 Usage Data clause was removed. |
| 2026-04-08 | Training exception further broadened: “Le Chat Pro” carve-out extended to also cover “Le Chat Teams.” |
Version History
2025-11-11
- Explicit AI clause: YES
- Non-explicit predatory: NO
- Flagged keywords:
output ownership
Clause: output ownership
You and your authorized users may provide input to the Mistral AI Products (“Input”), and receive output from the Mistral AI Products based on the Input (“Output”). […] To the extent permitted by applicable law, you (i) retain all ownership rights in Input and (ii) own all Output. We assign to you all right, title, and interest, if any, in and to Output that we may have.
Clause: unconditional no-training commitment (not separately concept-tagged)
We do not have access to the Infrastructure or Your Data and do not use Your Data to train our Models.
2025-12-15
output ownership unchanged from 2025-11-11 (renamed Input/Output terminology but same substance).
Clause: royalty-free (training license — new, reversing the 2025-11-11 commitment)
Mistral AI will not use Customer Content to train its artificial intelligence models except (a) when Customer or an End User provides Feedback to Mistral AI, (b) when Customer Content is flagged as part of Mistral AI’s automated moderation or reported by a user under the Additional Terms, or (c) as otherwise may be provided in an Order Form. Customer grants Mistral AI a perpetual, irrevocable, worldwide, non-exclusive, non-transferable […] royalty-free, fully-paid license […] to use Customer Content solely as provided in the preceding sentence to train Mistral AI’s artificial intelligence models.
Clause: competing model ban, output ownership restriction (new)
Customer may not use: (a) image Outputs to develop or train any image generation product that competes with a Mistral AI Product; or (b) Outputs generated using our web-search feature to (i) copy, store, archive, cache, or create a database of the Outputs, (ii) redistribute, resell, or sublicense the Outputs, (iii) as part of any machine learning or similar algorithmic activity, or (iv) to create, train, evaluate, or improve commercial products or services that Customer makes available to third-parties.
Clause: feedback (new)
Customer may provide feedback to Mistral AI in connection with its use of the Mistral AI Products (collectively, “Feedback”). By providing Feedback, Customer acknowledges and agrees that Mistral AI may use, copy, disclose, license, distribute, and exploit such Feedback, along with the Input and Output associated with such Feedback, in any manner without any obligation, royalty, or restriction based on intellectual property rights or otherwise.
Clause: ai disclosure (new)
Customer may not represent or imply that the Output was generated by a human when it was generated by the Mistral AI Products.
Clause: Usage Data (new, not separately concept-tagged — removed 2026-03-12)
Mistral AI may create aggregated and anonymized datasets (“Usage Data”) based on Customer’s use of the Mistral AI Products. Mistral AI may use Usage Data for its business purposes, including to conduct research, improve the Mistral AI Products, or enhance product performance, functionality, and user experience. Usage Data is the sole property of Mistral AI and does not constitute Customer Content.
2026-03-12
output ownership/feedback/ai disclosure unchanged from 2025-12-15.
Clause: royalty-free (training license — broadened)
Mistral AI will not use Customer Data or Outputs to train its artificial intelligence models except (a) for Customer Data or Outputs used or generated in connection with the use Mistral AI Products under a free subscription or Le Chat Pro where Customer has not opted-out of training, (b) when Customer or an End User provides Feedback to Mistral AI, (c) when Customer Data our Outputs are flagged as part of Mistral AI’s automated moderation or reported by a user under the Additional Terms, (d) as otherwise may be provided in an Order Form or (e) when Customer uses Labs Models.
Clause: Labs Models opt-out override (new, not separately concept-tagged)
By using Labs Models, you acknowledge that (i) Mistral AI may use Customer Data and Outputs generated from Labs Models to train its artificial intelligence models, unless you have activated zero data retention, and (ii) the opt-out preferences you selected for other Mistral AI Products does not apply to Labs Models. If you do not want your data or outputs used for training, do not use Labs Models.
Clause: competing model ban (narrowed — web-search Output restriction removed)
Output Restrictions. To the extent permitted by applicable law, Customer may not use image Outputs to develop or train any image generation product that competes with a Mistral AI Product.
The Usage Data clause from 2025-12-15 was removed entirely in this version.
2026-04-08
output ownership/feedback/ai disclosure/competing model ban unchanged from 2026-03-12.
Clause: royalty-free (training license — Le Chat Teams added)
Mistral AI will not use Customer Data or Outputs to train its artificial intelligence models except (a) for Customer Data or Outputs used or generated in connection with the use Mistral AI Products under a free subscription, Le Chat Pro or Le Chat Teams, where Customer has not opted-out of training […]