Consumer AI account or enterprise license: what the contract changes
An enterprise AI assistant and a consumer account can look alike on screen. The contract decides what happens to what you type.
On this page
In 30 seconds
- Where it happens
- When you ask a generative AI to correct, translate or summarize a text, tasks the CNIL lists among the uses of generative AI[2].
- The risk
- On a public service, the query is visible to the organization that provides the model, which stores it[3].
- The reflex
- Enter no confidential or personal data into a consumer service, and stick to the uses your company's policy allows[2].
What it is
Two AI assistants can look alike on screen and handle what you type in very different ways. The difference lies in the contract, not in the interface. When an organization uses Microsoft Copilot or Microsoft Copilot Chat, that use is covered by Microsoft's Data Protection Addendum (DPA) and the Microsoft Product Terms, with Microsoft acting as data processor[1].
The CNIL describes the same mechanism from the organization's side. For a system hosted by a third party, it asks for a processing contract with the host and, where applicable, with the system's provider. That contract must specify the scope of each party's responsibility and which accesses to the processed data are authorized[2]. Microsoft Copilot serves as the example here because its commitments are published.
Straight answers
What separates a tool provided by the company from a consumer account?
The contractual framework. For Copilot in an organization, prompts and responses fall under the same contractual commitments that, according to Microsoft, apply to emails in Exchange and files in SharePoint[1]. Microsoft writes that it will not use the organization's data other than on the organization's instruction[1].
For a public service, the NCSC (UK) wrote in 2023, about ChatGPT, that the query will be visible to the organization providing the model, that queries are stored and will almost certainly be used at some point to develop the service or the model[3]. For its part, the ANSSI observes that with most services, the data sent is collected and used by the provider to optimize the models[4].
Is what I type used to train the model?
For Copilot under enterprise data protection, Microsoft states that prompts, responses and data accessed through Microsoft Graph are not used to train foundation models[1].
For a public service, the NCSC stated in 2023 that a language model does not automatically add the information from a query to its model for other users, as of the time of writing (March 2023). It added that the provider, its partners or its contractors can read queries and may incorporate them into future versions[3].
Are my exchanges with the tool recorded?
With enterprise data protection, Microsoft states that Copilot Chat prompts and responses are logged. They are stored in Exchange, in the organization's Microsoft 365 tenant, for audit and eDiscovery[5]. The same retention policies as for Microsoft Copilot may apply to Copilot Chat[5]. The exact controls and policies vary with the underlying subscription[1].
What do the CNIL and the ANSSI say about entering data?
The CNIL recommends never sharing confidential information, such as personal data or company data, when using a consumer service[2]. It specifies that users should submit only information they are authorized to share[2].
The ANSSI takes the same line: sensitive data must absolutely not be sent to publicly accessible third-party generative AI services. It lists Diffusion Restreinte or classified data, personal data, the company's contractual, legal and financial data, and IT secrets such as passwords or API keys[4].
How it works
A processing contract does not change what the model does. It changes who answers for what. For Copilot, Microsoft describes commitments on encryption of data at rest and in transit, isolation of data between tenants and GDPR compliance[1]. It adds that Copilot respects the organization's identity model and permissions, inherits its sensitivity labels, applies its retention policies and supports auditing of interactions[1].
In Copilot Chat, Microsoft states that enterprise data protection is available at no additional cost. A green shield, displayed at the top of the interface next to the New chat button, shows that it applies[5]. Files uploaded to Copilot Chat are stored in the user's OneDrive for Business[5].
Where no local, secure and specialized system exists, the CNIL recommends determining to what extent the provider can reuse the data supplied, and adapting the use accordingly[6]. For a system used through an API, it stresses that control of the system lies almost entirely in the provider's hands. It therefore advises avoiding entering personal data into it as far as possible and paying particular attention to the contractual terms[2].
The CNIL also recommends that the organization frame the use of generative AI with internal policies or charters that define authorized and prohibited uses[2]. That charter says, in practice, which tool has been chosen and for which data.
What you do
- You stick to the uses your company's policy allows. The CNIL recommends that it define them, and that, depending on the deployment mode, it prohibit supplying certain confidential or personal data[2].
- In a consumer service, you do not enter sensitive data, or queries that would be a problem if they became public. This is the recommendation of the NCSC (UK)[3].
- You do not paste a password or an API key into a third-party AI tool[4].
- You review the answers before reusing them. The CNIL states that users should take a critical look at the outputs and check their accuracy[2].
Pitfalls
- A web search has its own rules. In Copilot, queries sent to the Bing service fall under the Microsoft Services Agreement and Microsoft's privacy statement. Microsoft acts there as an independent data controller[1]. According to Microsoft, these queries are not covered by the European Union Data Boundary (EU Data Boundary)[1].
- Agents have their own terms. Microsoft invites users to consult the agent's privacy statement and terms of use to find out how it handles the organization's data[1].
- Protections depend on the subscription. Microsoft writes that the exact controls vary with the organization's plan[1].
- A consumer account with a personal address. For non-confidential uses, the CNIL accepts a consumer service accessed with dedicated professional addresses, with appropriate safeguards. It sets two conditions: avoid accounts created with personal addresses and, where applicable, disable the provider's ability to reuse usage data[2].
- Terms of use nobody has read. The NCSC (UK) writes that the terms of use and privacy policy must be well understood before asking sensitive questions. It also cites the risk that queries stored online are hacked, leaked or accidentally made public, and that of the operator being bought by an organization with a different approach to privacy[3].
- Information entered that comes back out later. OWASP notes that users need to understand the risk of unintentionally providing sensitive data, which may then appear in the model's outputs[7]. It recommends that application publishers offer clear terms of use, with the option to opt out of having users' data enter the model[7].
Sources
- Microsoft, “Enterprise data protection in Microsoft Copilot and Microsoft Copilot Chat”. https://learn.microsoft.com/en-us/copilot/plan/enterprise-data-protection (accessed 14 November 2026)
- CNIL, “Les questions-réponses de la CNIL sur l’utilisation d’un système d’IA générative”, 18 July 2024. https://www.cnil.fr/fr/les-questions-reponses-de-la-cnil-sur-lutilisation-dun-systeme-dia-generative (accessed 14 November 2026)
- NCSC (UK), “ChatGPT and large language models: what's the risk?”, 14 March 2023. https://www.ncsc.gov.uk/blog-post/chatgpt-and-large-language-models-whats-the-risk (accessed 14 November 2026)
- ANSSI, “Security recommendations for a generative AI system”, 29 April 2024. https://messervices.cyber.gouv.fr/documents-guides/security_recommandations_for_a_generative_ai_system.pdf (accessed 14 November 2026)
- Microsoft, “Privacy and protections”. https://learn.microsoft.com/en-us/copilot/plan/privacy-and-protections (accessed 14 November 2026)
- CNIL, “Comment déployer une IA générative ? La CNIL apporte de premières précisions”, 18 July 2024. https://www.cnil.fr/fr/comment-deployer-une-ia-generative-la-cnil-apporte-de-premieres-precisions (accessed 14 November 2026)
- OWASP, “LLM02:2025 Sensitive Information Disclosure”, 10 April 2024. https://genai.owasp.org/llmrisk/llm022025-sensitive-information-disclosure/ (accessed 14 November 2026)
Cite this article
Brain, “Consumer AI account or enterprise license: what the contract changes”, Cyberpedia, 2026. https:/
BibTeX
@misc{brain_cp307_en_2026,
author = {Brain},
title = {Consumer AI account or enterprise license: what the contract changes},
year = {2026},
howpublished = {Cyberpedia},
url = {https://startbrain.io/cyberpedia/en/articles/consumer-ai-account-vs-enterprise-license-terms/},
note = {CC BY 4.0}
}