ETHICLAW
MANIFESTO
For verifiable governance of artificial intelligence
Trust cannot be demanded. It must be demonstrated. |
|---|
Artificial intelligence is entering decisions that affect people, organisations and institutions. The question is no longer only how capable a system is, but whether its behaviour can be understood, verified, challenged and corrected.
Today, values, decision criteria and responsibilities are often distributed across opaque systems. When a decision is wrong, it is difficult to establish which component contributed, which principle was applied, or how to intervene without changing the entire system.
EthicLaw was conceived to make governance a property of the architecture rather than a declaration of intent. It does not propose a universal set of values, claim to determine what is right, or replace human judgement. Instead, it separates the supervisory mechanism from the normative corpus it applies.
In EthicLaw, principles must be declared and versioned; modules must be observable, testable and replaceable; decisions must produce evidence; enforcement must be effective; and the audit must link what the system observed, evaluated, decided and executed.
Ethics is the first case study, not the limit of the project. In principle, the same infrastructure could apply legal obligations, professional codes, corporate policies, healthcare protocols or safety rules. Each domain remains responsible for its principles; EthicLaw aims to standardise how they are declared, applied, updated and verified.
Modularity does not eliminate complexity or automatically guarantee correct decisions. It does, however, make it possible to localise contributions, limitations, errors and responsibilities along boundaries designed in advance. Assurance must also be proportionate to the observability and actual intervention capabilities available.
EthicLaw’s ambition is not to create an arbiter of truth. It is to build infrastructure in which AI behaviour can be governed through reproducible evidence and subjected to independent oversight.
DO NOT STANDARDISE VALUES.
Standardise how their application can be verified.
Launched with the support of 3FLG