Reactor
A Governable Reference Architecture for Enterprise AI Agent Orchestration
A design for persistent agents, explicit authority, durable obligations and recoverable effects, with declared trust assumptions and validation criteria.
GOVERNANCE · AUTONOMY · INTELLIGENCEPUBLIC RESEARCH & REFERENCE DESIGNS
Intelligence
with integrity.
Powerful intelligence needs accountable systems.
We explore how to build them.
AEGISES is an initiative exploring the governance, autonomy
and ethical evolution of artificial intelligence.
Explicit authority.
Inspectable evidence.
01 / SELECTED RESEARCH
Ideas, made inspectable.
Read the argument.
Examine the architecture.
Follow the evidence.
Beyond the Bill Count
State AI Legislative Maturity Atlas
Six analytical diagrams and a source-linked ledger distinguish legislative design, implementation and outcomes. A representative corpus across the 50 states, with DC shown separately.
02 / ARCHITECTURES & TECHNICAL DOSSIERS
From policy to the machine.
Proposed architectures, implementation choices
and the engineering work between them.
POLICY COMPILATION
Typed Policy Reactor
A monitor-compiler design for lowering typed, default-deny policies into enforcement mechanisms across the stack.
Read dossierSOVEREIGN INFRASTRUCTURE
Sovereign Compute Architecture
A proposed capability-typed compute stack, with distinct machine roles and a staged path through changing hardware.
Read architecturePUBLIC POLICY ONTOLOGY
From Users to Registers
A proposed public policy ontology for CHERI, connecting human-facing policy with kernel, datapath and hardware enforcement.
Read the proposalENGINEERING GAP ANALYSIS
The Custom-Code Gap Map
An engineering dossier on CHERI–seL4–NixOS: reusable components, integration boundaries and proposed custom work.
Explore the map03 / OUR APPROACH
Autonomy should
be earned.
We bring policy questions and systems engineering into the same frame. The work is to make authority explicit, consequences attributable, and claims open to examination.
INTELLIGENCE WITH INTEGRITY
Define the authority.
Ask who may act, who may authorize, and where the boundary is enforced. Put those distinctions into the design.
Preserve accountability.
Connect decisions to responsible actors, durable records and paths for review and recovery.
Make claims inspectable.
Publish assumptions, sources and validation criteria. Distinguish a proposed design from an implemented mechanism and an observed result.
04 / PARTICIPATE
Help make intelligence
governable.
We welcome researchers, engineers, policy specialists and partners interested in the governance and ethical evolution of AI.
Bring a question, a critique or a concrete contribution.
contact@aegises.org Explore the AGI RFC roadmap