GIAS dimension
Control Surfaces · Governed Intelligence Architecture
A control surface is any point where an AI system can influence a decision, route information, invoke a tool, change memory, trigger execution, or alter what the organization believes to be true. Ananke makes those surfaces visible, governable, and auditable.
Control thesis
Trust boundaries define where authority changes hands. Control surfaces define where authority can be exercised. In GIAS, any surface that can change state, route context, trigger action, update memory, or shape belief must be explicit, governed, and reconstructable.
Why control surfaces matter
Input Surface
Where human intent enters the system and must be interpreted without inheriting unlimited authority.
Routing Surface
Where the system decides which model, provider, tool, memory store, or workflow receives the request.
Execution Surface
Where approved language becomes operational action through APIs, tools, databases, workflows, and external services.
Persistence Surface
Where the system decides what becomes memory, what becomes evidence, what is retained, and what must be forgotten.
Reference structure
Input control surface
Input is not merely text. It can carry commands, implied permissions, hidden context, injected instructions, file contents, or operational requests. The first control surface determines whether the system should treat the request as conversation, analysis, action, escalation, or threat.
Governance control surface
The governance surface determines whether a request may proceed, be modified, escalated, deferred, denied, or routed elsewhere. This is where Rita's reference architecture becomes enforceable.
Context surface
Memory retrieval, user profile, institutional policy, documents, previous decisions, and environmental signals must be assembled under governance.
Routing surface
Routing determines model, provider, tool, agent, memory store, or workflow based on risk, policy, cost, capability, and privacy requirements.
Inference surface
The model generates possibilities, but inference remains downstream from governance and upstream from review, execution, persistence, and observation.
Execution surface
Tool calls, API requests, database writes, notifications, external workflows, and system changes require explicit authorization.
Persistence and evidence surface
Persistence determines what becomes memory, what becomes evidence, what is retained, what is redacted, and what can later be reconstructed through forensic replay.
Control chain
A governed intelligence architecture does not rely on a single moderation check. It evaluates authority repeatedly as identity, context, governance, inference, execution, persistence, and observation move through the system.
1. Input
Classify intent and detect unsafe instruction patterns.
2. Context
Assemble only the context permitted for this user, role, and request.
3. Governance
Evaluate policy, risk, permissions, and escalation requirements.
4. Routing
Select the approved model, provider, tool, or workflow path.
5. Execution
Allow action only within authorized scope.
6. Evidence
Preserve the evidence required to reconstruct what happened.
Operational controls
Admin Policy Controls
Institutional administrators need policy surfaces for roles, routing, retention, approval, and escalation.
User Consent Controls
Users need clear control over memory, data use, sharing, deletion, and personalization.
Developer Tool Controls
Engineering teams need controlled access to model routing, tool permissions, telemetry, logs, and test replay.
Compliance Review Controls
Regulated organizations need review surfaces for approvals, denials, incidents, exceptions, and audit trails.
Runtime Intervention Controls
High-risk actions require the ability to pause, block, escalate, rewrite, or require human review.
Forensic Replay Controls
Leaders must be able to reconstruct the governed path that produced a consequential outcome.
Rita + Palladium + Origin
Rita implements the control-surface model defined by GIAS, translating authority, policy, risk, routing, execution, persistence, and escalation into operational decision logic.
Understand Rita
Palladium operationalizes control surfaces for organizations through policy configuration, runtime enforcement, model routing, tool permissions, evidence capture, incident review, and forensic replay.
Explore Palladium
Origin applies governed control surfaces to the individual: memory consent, identity continuity, privacy preferences, context boundaries, and user-owned intelligence that persists above model providers.
Explore OriginArchitecture pages
Next: Threat Model
Control surfaces show where AI can influence decisions, memory, routing, persistence, and action. The threat model defines what can go wrong at each surface — and how governed intelligence contains, records, and responds before harm scales.
