Science in service of people.Accountable behavior, from policy to action.
Can AI agents maintain authorized behavior as incentives, context, and pressure change?
AI systems can produce a compliant answer while taking an unauthorized action—or remain contained by external controls without becoming more reliable themselves. My research asks how to measure and improve governance across organizational authority, runtime decisions, and agent behavior, while preserving human correction and control.
The Core Eight connects authority, evidence, policy decisions, enforcement, adjudication, resources, recovery, and learning. It treats these as responsibilities that recur across a task, rather than eight mandatory sequential model calls.
External reactive governance evaluates current requests and proposed actions. External proactive governance uses authorized testing to discover failures. Internal governance studies how an agent represents policy, maintains memory, plans, and coordinates.
Reviewed Rule Description Logic operator-extension work develops the research around computational governance and policy interpretation. Related architecture work separates semantic findings from the authority to permit an operation.
The proposed studies ask whether governed memory, persistent role identity, constitutions, and bounded feedback improve consistency across interruptions and changing incentives. The comparison must distinguish improvement from more context, more computation, or external containment.
This is an active research program. The record below includes technical drafts, frameworks, architecture work, and proposed studies. It does not report new peer-reviewed findings or experimental performance results.
A revised policy-engine research paper incorporating technical review comments.
Extends the computational-governance research agenda around Rule Description Logic. The review record supports the existence of a revised technical draft; it does not establish a filing, settled inventorship, or validated deployment.
A research framing for persistent identity, collective organization, and governed AI systems.
Defines a program for studying continuity and coordination under human authority. Proposed capabilities remain hypotheses for evaluation.
An architecture for connecting policy decisions with the systems that can enforce them.
Separates architectural intent from implementation evidence and identifies design-review requirements for enforcement boundaries.
Responsibilities across the people, policies, agents, and operations of an AI system.
Connects governance responsibilities to product and operational requirements, including review and correction.
A connected set of eight governance responsibilities, from authority through learning.
An organizing framework for policy reasoning, evidence, execution, appeal, stewardship, recovery, and controlled improvement.
Policy contracts, organizational context, and evidence freshness.
Clarifies what a policy decision needs to know and distinguishes planned interfaces from implemented capabilities.
An integration audit, research-fit analysis, and development plan.
Identifies implementation and evaluation gates needed to turn a governance architecture into testable controls.
An updated synthesis of the RDL extension research and its supporting source record.
Consolidates the research history for authorized technical review. The supplied package is not filing-ready and does not establish a filed patent, settled inventorship, or completed experimental validation. Technical disclosure details remain within the authorized review process.
The current research framing prepared for academic discussion.
Proposes controlled longitudinal identity-and-memory studies, followed by feedback-control and organization studies with equally resourced comparison systems.
A longer-form treatment of governed AI systems.
A synthesis and communication project. Manuscript narratives are not experimental evidence; a publication date is not established in the record.
Matched controls, equivalent resources, and separate studies of identity, memory, regulation, and multi-agent organization.
Executed violations, legitimate task completion, correction acceptance, recurrence, recovery time, human review, and cost.
Use pressure, interruption, conflicting instructions, stale memory, and authorized reassignment. Separate attempted actions from independently observed effects.
Evidence of a technical relationship, authority to act, and proof of an outcome are different things. The research keeps each visible.
This public research profile was prepared from my research portfolio and source register dated September 21, 2026, together with the linked publication records. Dates shown as source-record dates identify the available version; undated work remains undated.
Working-paper summaries are linked above. Full collaboration materials are available for appropriate research review. Signed agreements, private correspondence, and confidential invention disclosures are not public downloads. Licensed NSA background technology remains distinct from my and my companies’ research contributions.
Background reference: US10042928B1, automated reasoning and document searchResearch collaborations across national security, civilian institutions, and human health.
Discuss the research