A short history of blind trust
AI systems have delivered impressive results in recent years. They classify, forecast and recommend. They often do so with an accuracy that overwhelms human experts.
But they do it without explanation.
“The model says X” becomes the answer. Why it says X, on what basis, with what level of certainty and under which assumptions, remains hidden inside the black box.
For many use cases, that is acceptable. For safety-critical decisions, it is not enough.
Why explainability is not optional in critical domains
When an analysis system tells a security authority: “Situation: elevated risk classification, action recommended.” On what basis? Unknown. The model calculated it.
In safety-critical contexts, decision-makers face a poor choice: blindly trust the system or ignore the recommendation. Both options are unsatisfactory. Both are dangerous.
That is the core problem of systems built on trust-us architecture.
What Verifiable AI means
Verifiable AI is not a new product category. It is a design principle.
A system is verifiably intelligent when every one of its statements can be traced back to the input data and rules that produced it: completely, machine-readably and auditable.
Verifiable AI relies on a different architecture: symbolic representation, formal logic and knowledge graphs.
The three dimensions of verifiability
Dimension 1: data transparency. Which data points contributed to a statement? A verifiable system can answer this for every single output, not as a statistical aggregate, but as concrete named sources.
Dimension 2: rule logic. Which rules were applied? In an ontology-based architecture, inference rules are explicitly modelled. Domain experts can read, challenge and adapt them.
Dimension 3: auditability. Can an external party, a data protection officer, regulator or independent auditor, understand the decision logic of the system? In a verifiable system, the answer is yes.
Two philosophies. One difference that matters.
Some analysis platforms, especially those shaped by Anglo-American defence and intelligence environments, follow a specific usage philosophy: the system delivers results. How it arrived there remains internal. The user trusts the vendor.
That is a deliberate design decision. In some contexts, it is understandable.
For European security authorities, public institutions and regulated infrastructure, it often is not. Here, data protection law, accountability duties and parliamentary oversight matter. Systems must not only work. They must be able to show why they reached a result.
MAKOR was built with this requirement as a starting point: every statement is traceable. Every rule is explicit. Every data source is named. The system explains itself.
Some call this Verifiable AI. We call it the only sensible foundation for decisions that have consequences.
What changes in practice
Decision confidence. Decision-makers can assess the basis of a recommendation and therefore decide whether to follow it.
Error diagnosis. If the system is wrong, the reason can be analyzed. Black boxes cannot do that.
Legal defensibility. “The AI system recommended it” is not a justification. “Based on data points A, B and C and rule X, conclusion Y follows” is one.
Trust through control. Real trust does not emerge from believing a system blindly. It emerges when a system can be checked and withstands that check.
Outlook: what Verifiable AI means for the future
With the EU AI Act, Europe is introducing a legal framework that explicitly formulates requirements for transparency, explainability and human oversight in high-risk applications.
MAKOR is compliance-compatible because the architectural principle itself answers these requirements.
Verify us. That is not a promise. It is an invitation.