What is being proposed?
Action, actor, target, requested scope and the system that produced the proposal.
Architecture / Runtime governance
Make autonomous action
inspectable.
UBIKITOS is a governance boundary between an AI proposal and a consequential action. The architecture keeps authority explicit, evidence attached and the final decision reconstructable.
01 / The position
A boundary, not
another model.
Agents can plan, call tools and produce confident outputs. None of that answers whether a particular action should happen now. UBIKITOS evaluates the proposal in context, then returns a governed outcome: allow, defer, escalate or block.
The model, orchestration layer and business systems remain free to change. The boundary is the stable place where a consequential proposal meets policy, evidence and human accountability.
02 / The control plane
Five parts.
One trace.
What supports it?
Current context, provenance and the evidence required by the workflow before action.
Who permits it?
Explicit, scoped permission: who may act, on what systems, under which conditions.
What changes?
Reversibility, uncertainty and the cost of a wrong action determine the control level.
Why did it happen?
Inputs, policy version, evaluator result, escalation and outcome preserved for review.
03 / Example trace
From proposal
to proof.
A useful trace is compact enough for a system to emit and complete enough for a person to inspect later. The shape below is illustrative; a live deployment would bind each field to its approved policy and evidence sources.
Decision trace / refund-001
proposal.action = issue_refund
proposal.amount = EUR 180
evidence.order = verified
authority.mandate = refund-v1.0 (max EUR 200)
consequence.level = reversible / medium
decision.outcome = ALLOW
decision.reason = evidence + authority satisfied
record.created_at = 2026-09-05T12:00:00ZDesign test: if a reviewer cannot explain the outcome from the preserved trace, the system is observable but not yet governable.
04 / Continue