What governance issues can an AI operating system not solve?
Many believe that whoever introduces a system that monitors everything has everything under control.
This belief is dangerous.
Five governance problems that no technical system can solve:
Responsibility cannot be configured.
Who is liable if an AI-powered credit decision is wrong? Who bears responsibility for a discriminatory algorithm in HR? These questions are organisational and legal – no platform answers them.
2. Risk classification requires judgment.
Whether a system is to be classified as high-risk under the EU AI Act depends on the scope of application, intended purpose, and deployment context. This is a legally structured and technically substantiated classification – not a technical parameter.
3. Ethical dilemmas demand leadership decisions.
Efficiency versus data protection. Automation versus employment. Personalisation versus manipulation. These trade-offs cannot be delegated to configuration settings.
4 Third-party providers remain a black box.
Anyone using Microsoft Copilot, Salesforce Einstein or an external applicant management tool is using external models with their own training data and error rates. An AI OS sees the output - not what happens in the models.
5. Strategic prioritisation is a management task.
Which processes should be AI-assisted? Where does AI make economic sense and where does it merely create complexity? These questions should be asked before implementing a platform, not after.
—
An AI operating system is a control tool. It presupposes that the company already knows what is to be controlled – and according to what standards.
Anyone who doesn't meet these prerequisites is buying a governance problem, not a governance tool.
And an AI OS generates new governance risks itself. Which ones? More on that in Part 3.
The five biggest governance gaps with AI OS? Responsibility, risk classification, conflicts of interest, black box, prioritisation.
Author: Achim Korten, April 2026