AI providers’ terms: training, retention, location

Regulatory briefing

Asmaa EL FORKANI 4 min read

The terms of AI providers almost all distinguish two regimes: that of consumer offerings and that of business offerings. Reading them against four criteria (training, human review, retention, place of processing) makes it possible to compare offerings without ranking them. It also makes it possible to check whether a tool is suitable for a given use.

ASKonsult – Strategy & Operations

The text: four criteria for consumer and business offerings

  • Training: are exchanges used to improve models, by default or on an opt-in basis?
  • Human review: can people read the exchanges, and under what conditions?
  • Retention: how long are exchanges retained, including after deletion?
  • Place of processing: where are the data processed, and can the organisation choose it?

The findings below result from the providers’ pages consulted on 21 October 2026. However, providers frequently amend these terms. The references therefore give the links for verification at the date of use.

Consumer offerings: training, review and retention

OpenAI states that it may use content from its consumer services to train its models. The user may, however, opt out through a dedicated setting. Google states that human reviewers examine a portion of Gemini Apps exchanges. It also states that it retains reviewed conversations for up to three years. Anthropic provides, for its Free, Pro and Max plans, for use in training if the user allows it. In that case, retention extends to five years, compared with 30 days otherwise.

Mistral AI states that it uses data from the individual Vibe plans and from the free mode of Mistral Studio for training. Users may nevertheless opt out at any time, whereas the provider excludes Vibe Enterprise customers by default. In addition, OpenAI and Mistral AI both address feedback on a response, whether approval or disapproval. Such feedback, they specify, may make the exchange concerned usable for training.

These regimes thus share one characteristic: the choice belongs to the individual user, not to the organisation.

Business offerings: training, retention and data location

OpenAI states that by default it does not use data from its business offerings and its API for training. OpenAI may nonetheless retain API inputs and outputs for up to 30 days. For certain eligible uses, the customer may also request zero retention. Google, for its part, states that in Google Workspace it does not use customer content to train models outside the customer’s domain without permission. Google adds that humans do not review that content.

Microsoft states that Microsoft 365 Copilot prompts and responses are not used to train foundation models. It adds that European Union customers benefit from data location commitments within the Union. However, those commitments currently exclude the Anthropic models supplied as a subprocessor. Finally, Anthropic excludes from the regime applicable to consumers the services governed by its commercial terms. These services include the API and enterprise offerings.

The place of processing remains the least documented criterion on public pages. For business offerings, the organisation therefore checks it in the data processing agreement and the list of sub-processors. Where applicable, it also checks the data location options offered.

Implications of consumer and business offerings in Morocco

For a Moroccan organisation, the reading grid serves two purposes. First, it determines admissible use. A tool whose exchanges the provider may use for training or review receives no personal data or confidential information. Second, it determines the formalities. Indeed, the place of processing conditions the analysis of the transfer under Articles 43 and 44 of Law No. 09-08. That analysis also relies on the list of States drawn up by the CNDP.

The comparison of consumer and business offerings does not, however, designate any preferable provider. The right choice depends on the use, the data involved and the contractual safeguards obtained.

Points of attention: consumer or business offerings

  • Distinguish, for each tool, the offering used (consumer or business).
  • Record and date the four criteria for each provider.
  • Obtain the data processing agreement and the list of sub-processors for any business offering.
  • Review the grid whenever the provider’s terms change.

References

See also: Free AI tools: what users pay in data; Grid – Assessing an AI provider. Information Systems & Digital Transformation practice.

General information document; it constitutes neither legal advice nor an opinion on a particular situation. References to legal texts should be checked against the version in force on the date of use.

Consulting firm not registered with any Bar; no representation before the courts and no act reserved to the regulated legal professions.