One employee pastes a contract into a chatbot to get a summary. Another has it draft a reply to an unhappy customer, copying in the history of the exchanges. A third gives it an Excel file to “highlight the best customers”. None of this is malicious: it is simply efficient. But in each of these cases, company information has just left its information system.
The aim of this article is not to make you give up on AI. Used well, it saves hours every week. It is about knowing where the risks are, what the Moroccan framework says, and which simple rules to put in place.
1. What you paste into a chatbot leaves the company
When you use an online AI assistant, the text you type and the files you upload are sent to the provider’s servers, most often outside Morocco. Depending on the plan chosen and your settings, conversations may be retained, viewed in some cases, or even used to improve the models. The rules are not the same for a free consumer plan and a business plan.
The most exposed information is also the most common in an SME:
- customer files, contact details, order histories;
- ID documents, bank details (RIB), contracts, personnel files;
- health data, in medical practices and clinics;
- passwords, access keys, source code containing secrets.
2. The Moroccan framework: Law 09-08 and the CNDP
In Morocco, Law No. 09-08 governs the processing of personal data, under the supervision of the CNDP (the National Commission for the Control of Personal Data Protection). Processing the data of your customers or employees entails obligations: a defined purpose, informing the individuals concerned, and appropriate security measures.
An often-overlooked point: the transfer of personal data abroad is regulated. Depending on the destination country and the nature of the data, it may require authorization from the CNDP. Yet submitting a customer file to an AI hosted outside Morocco is precisely such a transfer.
This article provides practical guidance and does not constitute legal advice. For a specific processing operation, contact the CNDP or your legal advisor.
3. Where AI really pays off, without risk
The good news is that the most profitable uses require no sensitive data:
- writing and rewording non-confidential texts: ads, product sheets, social media posts;
- translating marketing content between French, Arabic and English;
- summarizing public documents: legal texts, calls for tenders, technical documentation;
- programming assistance, provided you never paste passwords or keys;
- analyzing anonymized data: names, numbers and identifiers are removed before sending.
4. “Open source and installed in-house”: a good idea, provided you check
For sensitive data, an attractive option is to run the AI on your own servers: the data no longer leaves. But you still have to trust the software.
We checked this recently for our own use. Before entrusting anything to an open-source voice synthesis studio, we audited it. Good news: no malicious behavior. But three findings needed fixing:
- its interface accepted commands from any website open at the same time in the browser, without any authentication;
- it included an automatic update feature capable of downloading and installing new code;
- several libraries it used had known vulnerabilities.
We isolated it before any use: dedicated account without privileges, outbound network access cut off, write access limited to its own folder, access restricted to the local interface. The exposure score measured by the system dropped from 8.5 to 2.9 out of 10.
The lesson applies to all tools: open source does not mean risk-free, and local does not mean isolated. An audit and isolation are essential before entrusting it with your data.
5. Check licenses before publishing
Synthetic voices, images, music: the free plans of AI tools often prohibit commercial use. An advertising video made with a “free” voice may therefore breach the terms of service. Before publishing, read the terms, or opt for a business plan or a model under a permissive license (MIT, Apache-2.0).
And one absolute rule: never clone a person’s voice or image without their written consent.
6. The 7-point checklist
- Take inventory of the AI tools actually used in the company: there are often more than you think.
- Write a short usage policy: what can be submitted, and what is never submitted.
- Opt for business plans, with named accounts, two-factor authentication and a commitment not to reuse data.
- Anonymize before sending: names, national ID (CIN) numbers, phone numbers, addresses.
- For sensitive data, use hosted tools under your control, audited and isolated.
- Check the licenses of any generated content before publishing it.
- Train your teams: a short session is enough to establish the right habits.
How TAWSA supports you
Based in Marrakech, we help Moroccan businesses benefit from AI without exposing their data: inventory of uses, usage policy, selection and deployment of secure tools, audit and isolation of solutions installed in-house, and encryption of sensitive data with URAEUS. These engagements are part of our managed IT services offering.