What information is actually required?
Reduce the input to what the task needs. If the work can be done without a name, identity number or a full document, that information stays out.
Information security and responsible AI
Security starts with the cost of a mistake. We set what data enters, who may reach it, where a person reviews before an action, and how a vendor and a tool are chosen.
Security starts with the cost of a mistake
The same tool can fit one task and be dangerous in another. The work checks the information, the user, the action and what may happen if the result is wrong.
Who needs to see and act?
Access is given by role and purpose. Reading, editing and sending are not the same permission, and not every user needs all three.
Where is the result reviewed?
An AI output can be incomplete or wrong. The control point is set according to risk, before the result is sent, recorded or used to trigger an action.
What happens when the system is not sure?
Define when to stop, ask for more information, pass the work to a person or mark that the action was not completed.
Working with sensitive material
Sensitive information is not entered into a public tool without approval. As needed, the work uses anonymous material, an organizational AI environment, reduced permissions or sample data prepared in advance.
Choosing a vendor and a tool
The type of information, terms of use, storage location, management capabilities and the existing policy in the organization are examined. There is no single tool that automatically fits every client and every process.
Confidentiality and use of information
Information received for a project is used for the agreed purpose. The scope of access, the way of working and the responsibility are set in the engagement and according to the systems involved.
Retention and deletion
What is kept, where and for how long is set by the need of the project, the client's policy and the limits of the services used. When information is no longer needed, the agreed terms are followed.
Reviewing outputs
Checking an answer is not one action. In a document it is possible to return to the source. In a calculation the data and the formula need to be checked. In a system action, permission, destination and result need to be confirmed. Control is fitted to the kind of error that may occur.
What this page does not claim
These are working principles, not a statement of a standard, regulatory compliance or legal fitness. In each project the requirements and controls are defined according to the context and the people responsible in the organization.
AI news
A conversation about one work process
The first conversation describes the process, where the information lives and what needs to change. From there it is clear whether to start with training, automation or a system.