Our Work
Structured Governance Infrastructure for Institutional AI
Governance Architecture Model
SILT's governance infrastructure is organized across three discrete layers. Each layer addresses a distinct institutional requirement — from how a system is characterized, to how it is tested, to how it is supervised in production. Together they form a coherent governance stack for AI operating in regulated environments.
Layer 1Structural Classification
Layer 2Evaluation Environments
Layer 3Operational Oversight
Structural Classification Frameworks
Before an AI system can be governed, it must be characterized. SILT's classification frameworks provide architecture-level descriptors that translate system properties into governance-relevant signals — enabling proportional oversight rather than blanket policy.
Architecture-level descriptors
Formal characterization of system structure, scope, and capability boundaries.
Persistence and autonomy characteristics
Systematic assessment of memory, continuity, and degree of autonomous operation.
Governance calibration signals
Derived indicators that inform oversight intensity and reporting requirements.
Change-trigger definitions
Criteria that prompt reclassification when system behavior or deployment context shifts.
Example application: Autonomy & Continuity classification models for persistent AI agents operating across sessions in professional workflows.
Evaluation Environments
Structured evaluation requires controlled conditions — not ad hoc testing. SILT designs and operates assessment environments that produce reproducible, auditable evidence about AI system behavior under defined conditions.
Multi-domain assessment structures
Evaluation frameworks spanning the operational domains relevant to the deployment context.
Adversarial stress testing
Systematic exposure to edge cases, boundary conditions, and failure-mode scenarios.
Blind scoring methodologies
Separation of evidence collection from interpretation to prevent evaluator bias.
Version-controlled evaluation criteria
Criterion sets maintained under version control with audit trails for change management.
Transcript preservation
Full retention of evaluation sessions for downstream review, dispute resolution, and reassessment.
The Sentience Evaluation Battery (S.E.B.) is one environment within this layer — a structured assessment infrastructure for AI systems exhibiting high-continuity or autonomy characteristics. It is designed as governance infrastructure, not as a philosophical determination.
Oversight & Supervision Models
Governance does not end at deployment. SILT designs operational supervision layers that embed accountability directly into production workflows — enabling ongoing monitoring, escalation, and audit without disrupting institutional operations.
Output monitoring layers
Continuous review of AI outputs against policy-defined acceptability criteria.
Policy-sensitive flagging
Detection of outputs that require human review based on domain-specific risk parameters.
Audit logging systems
Immutable records of AI decisions, actions, and escalations for regulatory and internal review.
Escalation protocols
Defined routing for flagged outputs — to human reviewers, compliance teams, or automated holds.
Workflow integration
Supervision architecture embedded within existing institutional processes, not bolted on.
Standards Alignment
SILT's frameworks are designed to operationalize governance intent embedded in emerging international standards — translating regulatory requirements into institutional infrastructure that can be implemented, audited, and maintained.
EU AI Act
Risk-based classification and high-risk system obligations.
NIST AI RMF
Govern, Map, Measure, Manage functions across the AI lifecycle.
ISO AI Management Systems
Organizational frameworks for responsible AI development and deployment.
Institutional Deployment Model
SILT applies its governance stack across a structured institutional lifecycle. Each phase produces documented outputs that feed into the next, creating a continuous and auditable governance record.
01
System Inventory
Cataloguing all AI systems in scope with their operational contexts.
02
Classification
Applying structural classification frameworks to each system.
03
Baseline Evaluation
Running controlled evaluation environments to establish a documented behavioral baseline.
04
Oversight Calibration
Designing and deploying supervision layers proportional to classification outputs.
05
Reassessment
Scheduled and trigger-based re-evaluation as systems or contexts evolve.
