An engineering firm prototypes an AI design process on a cloud compute cluster: engineers send parameters, an AI algorithm uses the firm's experience database to compute optimal part shapes, and the CAD results go to production. What needs protecting?

The intellectual property (the experience database and the know-how in the algorithm), the calculation results (CAD drawings for parts), the IT resources on-premises and in the cloud, and the production plant. Algorithm and data must also be protected against manipulation (sabotage), and the AI itself must be kept under control.

The case: high-tech company S makes injection-moulded parts with optimised flow behaviour. Parts used to be designed in CAD from an experience database, and the CAD data goes straight into production. A small engineering team built a prototype where flow conditions (gas or liquid, speed, temperature, pressure, radiation…) are entered as parameters and an AI algorithm computes the best shape. Because it is very compute-heavy, one engineer rented a compute cluster in the cloud. The prototype works so well that management wants it in production.

Protection is not only confidentiality. Because the results flow directly into the production plant, a manipulated algorithm or dataset would cause sabotage by producing faulty parts, so integrity matters just as much.