Accountability
Named owners, decision rights, escalation paths, approval thresholds, and board-level oversight.
Research area / AI Governance
Governance that can be observed, evidenced, and improved.
We study how AI governance becomes an operating system: inventories, decision rights, controls, evidence, testing, monitoring, incident response, assurance, and executive accountability.
From policy to operating control.
Effective governance connects institutional accountability to the technical and operational systems where AI is selected, configured, deployed, monitored, and changed.
InventoryClassifyOwnAssessControlEvidenceApproveMonitorRespondReview
Control architecture
A governance system for real deployment.
We organize governance around control points that can be owned, evidenced, tested, and improved.
Risk classification
Consistent methods to determine impact, criticality, data sensitivity, autonomy, and control depth.
Evidence
Durable proof that required reviews, tests, approvals, monitoring, and remediation actually occurred.
Runtime control
Identity, authorization, monitoring, action boundaries, human intervention, and revocation for AI systems and agents.
Assurance
Independent review of whether governance mechanisms are designed effectively and operating as intended.
Change
Versioned controls and reassessment as models, workflows, data, regulations, and system capabilities evolve.
Agentic AI
Govern the action, not only the output.
As AI systems gain tools, memory, credentials, and authority, governance must extend into runtime behavior. Institutions need to know what an agent was allowed to do, what it actually did, which resources it touched, and who remained accountable.
Explore AI identity researchGovernance evidence
Controls should leave a trace.
Governance becomes operational when an institution can show what was reviewed, by whom, against which criteria, with what result, and what changed afterward.
Evidence chain
Design evidence
Policies, control objectives, risk methods, ownership, thresholds, technical patterns, and approval criteria.
Operating evidence
Test results, approvals, logs, monitoring, exceptions, human interventions, incidents, and remediation.
Assurance evidence
Independent challenge, findings, residual-risk decisions, reassessment, and version history.