Turns trust scoring from a display surface into an autonomy-control algorithm.
Abstract
This paper proposes an evidence-weighted autonomy ladder for AI agents, where trust events grant, narrow, pause, or escalate agent scope inside an Agentic OS.
Trust scores become operationally valuable when they affect future autonomy. This paper proposes an autonomy ladder controlled by a Trust Kernel: a runtime-adjacent layer that reads pacts, tool receipts, evaluator verdicts, and mission outcomes to adjust the agent's next scope.
These papers are built from the same trust questions Armalo is turning into product surfaces: pacts, trust oracles, attestations, and runtime evidence.
Teams will trust autonomous agents faster when autonomy is represented as a ladder with visible evidence and downgrade rules rather than a binary allow/deny permission model.
Measurement Plan
Measure conversion from agent setup to first governed workflow, operator approval rate, scope-expansion rate, incident downgrade rate, and buyer confidence in the audit report. Compare trust-kernel copy against generic trust-infrastructure copy.
Limitations
This is an operating model, not a universal safety guarantee. The ladder should complement, not replace, conventional security controls, tenant isolation, human approval, and incident review.
The Trust Premium: Platinum Agents Score 9.4× Higher Than Unranked Peers in Production