Imagine a hospital in Libreville wants to use AI (artificial intelligence) to summarize patient records before a medical meeting. Obvious time-saving. But a concern blocks the project: will the health data sent to the tool remain stored with the provider? Who will be able to consult them?
The same question arises in a bank analyzing customer complaints, an administration processing social requests, or a university responding to its students. AI can help, but only if the organization knows precisely what it shares, with whom, and for how long.
Recent articles
Excel with AI: how to maintain reliable results?
Can business software steal your sensitive data?
Can AI enhance the reliability of accident investigations?
Certificate
Master of Science (MSc) in Artificial Intelligence and Data Science
What is it, concretely?
When a company uses generative AI, it sends a prompt and receives a response: summary, analysis, draft letter, information extraction. The risk arises when this prompt contains sensitive data, such as a medical record, a payslip, or a customer complaint. OpenAI claims that its Private Safety Processing system can detect patterns of dangerous use across multiple exchanges, without giving OpenAI employees access to detailed content. The main challenge is therefore the balance between security and privacy. Good news: an organization can significantly reduce risks if it chooses the right contract, limits the data transmitted, and maintains human validation.
Concrete case: what to do and what not to do
Questions to Ask Before Acting
- Do the data sent to the AI contain personal, medical, financial, or HR (human resources) information?
- Does the provider keep the instructions and responses? If so, for how long?
- Can the data be used to train or improve the model, unless explicitly agreed?
- Does the contract provide for deletion, auditing, encryption, and precise access management?
- Do the data remain in a geographically acceptable area for your organization?
- Who validates the response before it is sent to a client, patient, student, or citizen?
- What procedure do we follow in case of an error, leak, fabricated response, or unauthorized use?
French 
English 