Trust & AI
Public AI must be accountable
When an algorithm informs a decision that binds the State, that decision must remain traceable, attributable and contestable. That is our Jazari ROOT vision, aligned with Morocco Digital 2030.
A model that cannot be audited has no place in a public decision.
Accountability before performance
Public administrations are adopting algorithmic assistants: scoring, file triage, decision support. The efficiency gain is real — so is the risk: who answers for a decision nobody can explain?
Our position is simple: in the public sector, accountability comes before performance. Every algorithmic recommendation must be logged with its inputs, model version and context; every final decision must carry the signature of an authorised human agent.
It is the same standard of proof as for classic administrative acts — extended to AI-assisted decisions, with the same sovereign building blocks: signature, timestamping, verifiable log, hosted on-premises.
Jazari ROOT: four non-negotiable principles
Named after Al-Jazari, the brilliant 12th-century engineer: public AI worthy of inventors who measured their mechanisms. ROOT — Responsibility, Openness, Opposability, Traceability.
Responsibility
Every AI-assisted decision has an identified human owner who signs it and can answer for it before a third party.
Openness
Model criteria and limits are documented and communicable to oversight bodies — no black box held against citizens.
Opposability
The decision log is signed and time-stamped: it stands as proof in case of appeal, years later.
Traceability
Inputs, model version, execution context: every recommendation can be reconstructed identically.
Our Jazari ROOT roadmap
Concrete building blocks, delivered in the order the public sector needs them.
-
1
AI decision log
Record every algorithmic recommendation — inputs, version, context — in the existing sovereign proof log.
-
2
Signature of the final decision
The authorised agent validates and signs: AI proposes, the human commits. Non-repudiation applies to the decision, not the model.
-
3
Model register
Versions, training data, evaluations and documented limits: the AI counterpart of the processing register.
-
4
Audit & counter-expertise
Replay a decision identically for oversight bodies: contestability becomes a feature, not litigation.
Is your administration preparing an AI project?
Let's frame its accountability before it goes to production.