Identity & Representation
Define persistent machine actors, ownership, provenance, model lineage, and the institutional context in which an AI system operates.
AI Governance
Governance for systems that can reason, act, and change.
Paris AI Organization researches the architectures of accountability required when advanced AI moves from producing outputs to exercising delegated authority across real systems.
Research thesis
AI governance is becoming a systems-engineering discipline.
Principles remain essential, but increasingly autonomous systems require controls that operate continuously: persistent identity, bounded authority, verifiable evidence, contextual authorization, monitoring, escalation, and intervention.
We study governance as an architecture that connects technical controls with institutional responsibility.
Governance architecture
From policy intent to runtime control.
Our program examines how governance can remain effective as models become agents, workflows become autonomous, and decisions are distributed across machines and people.
Authority & Permission
Translate human and organizational authority into machine permissions with purpose, scope, duration, conditions, and revocation.
Assurance & Evidence
Develop evidence structures for model behavior, system controls, data provenance, evaluations, incidents, and consequential actions.
Oversight & Intervention
Design monitoring, escalation, human review, rollback, and containment mechanisms that operate while systems are active.
Research questions
What must an institution know before it can trust an intelligent system to act?
How should AI identities be issued and revoked? Which actions need fresh authorization? How should an agent lose authority when context changes? What evidence is sufficient to reconstruct why a machine action occurred?
How should model risk, tool risk, data risk, and workflow risk combine? When should oversight be automated, when should it be human, and how can governance remain legible under real-time operating conditions?
Research direction
Build governance that scales with capability—not after it.
Machine-speed controls
Move governance from periodic review toward continuous evaluation of context, authority, behavior, and material change.
Evidence by design
Make auditability and provenance part of system architecture rather than an after-the-fact reporting exercise.
Human authority preserved
Design clear boundaries for delegation, escalation, intervention, and revocation when systems operate with increasing autonomy.
Selected research
Governance publications.
Governing Delegated Authority in AI Agents
A governance architecture for identity, bounded authority, escalation, and accountability when autonomous software acts on behalf of people and institutions.
Machine-Speed Governance: Continuous Controls for AI Systems
Why evidence, policy evaluation, monitoring, and control validation must become continuous as AI systems change faster than traditional review cycles.
Synthetic Data Assurance: Provenance, Privacy and Evaluation
A control model for deciding when generated data is fit for training, testing, simulation, analytics, and high-stakes evaluation.
Research collaboration
Advance the science of governable AI.
Paris AI Organization collaborates on governance research involving agents, assurance, evaluation, accountability, standards, and operational controls.
Collaborate ↗