Zero trust for AI systems
AI adds new things to protect: training data, models, prompts and, increasingly, the actions systems can take. Zero-trust principles apply to all of them.
Read article"A person can always override it" is not a safety strategy. Real human oversight has to be designed into the workflow.

Almost every AI proposal now includes the phrase “human in the loop”. Too often it means a person somewhere could, in theory, step in. In practice that person is overloaded, the system moves faster than they can follow and the override button is three screens away.
For mission-critical work, oversight has to be deliberate. Here is how we think about it.
Not every action deserves a human decision. Flagging a record for review, drafting a summary or sorting alerts can safely be automated. Locking down a site, dispatching a team or changing a patient’s care plan cannot. Draw that line explicitly, write it down and enforce it in the system, not in a policy document.
AI should complete routine work and prepare important decisions. People should make them.
An approval is only meaningful if the person approving can see why the system is asking. Every recommendation should come with its evidence: the source document, the video clip, the sensor readings and the confidence behind it. Answers that cannot show their sources should not be trusted with decisions.
This record is not bureaucracy. It is how an institution learns which recommendations were right, improves its models and answers to auditors, courts and citizens.
Finally, respect the limits of the person in the loop. A system that raises a hundred alerts an hour has not kept a human in the loop. It has buried one. Good design means fewer, better alerts, prioritised by risk and delivered to the right person on the right device.
Want to explore how this applies to your organisation? We share our approach with qualified organisations in a private briefing.

AI adds new things to protect: training data, models, prompts and, increasingly, the actions systems can take. Zero-trust principles apply to all of them.
Read article
Scaling AI without a management system is like scaling a factory without quality control. Standards such as ISO/IEC 42001 give institutions a practical framework.
Read article
Twenty months of writing about data, AI and sovereignty point to one idea: an intelligence layer that turns what an enterprise already owns into foresight, under its own control.
Read article