The organization must implement appropriate technical and organizational measures based on previous assessments. Some measures that may be relevant for the implementation of AI systems include:
- Technical adaptations and integrations: Existing systems must be adapted to the new AI system and new workflows. The organization may consider implementing constraints in the AI system to ensure it is used within the scope for which it was developed and validated [129]. An example could be an AI system that has been developed and validated only for adults. If a user inputs data for a child and age is included in the input, the AI system could notify the user that applying the system to this patient falls outside its intended use.
- Training: Before the system is put into use, the organization must ensure that relevant user groups receive training and understand the AI system’s functions, limitations, and ethical considerations. Users must be aware of the applicable incident reporting procedures, including which authority or reporting system to use if errors or incidents related to the AI system are identified during use, to ensure that such issues are detected and addressed. Existing incident management systems within the organization should be used as the basis for this [130]. This should also ensure compliance with the AI Regulation’s requirement for human oversight.