RESEARCH · DETERMINISTIC GOVERNANCE

AI Governance & Assurance

AI Governance & Assurance Research

Scientific and engineering questions at the boundary between AI proposals and consequential execution.

My research investigates the conditions under which AI-enabled workflows can offer sufficiently well-grounded assurance about their actions and outcomes. It emphasizes runtime authorization, independently evaluated evidence, verifiable state changes, and explicit limits of inference.

Research themes

Deterministic execution-layer governance; governance compilation; long-running agentic actions; proposal precision; conformance and admissibility; sequential AI inference and conditional assurance; and post-commit evidence.

A critical distinction

Schema conformity does not imply governance admissibility, and governance admissibility does not imply factual correctness. These are separate evaluation dimensions with different evidence requirements.

Method and evaluation

A useful evaluation protocol measures the precision of proposals, workflow completion, observed outcomes, governance conformance and coverage, and the operational overhead of enforcement.

The governance model, the AI or agent proposing actions, the runner or harness, and the independent execution control have separate responsibilities and fault modes.

Current research posture

Published research, public specifications, manuscripts, software experiments, and planned evaluations should be identified by their actual maturity and review status.