ModulesSAI-280
SAI-280 table of contents
Governed Agents and Controlled Tool Execution
Constrain agent tools, identity, memory, approvals, execution, and signed action evidence.
Learning outcomes
What you should be able to do
- Define an allowlisted tool registry
- Design human approval and confinement boundaries
- Produce verifiable run receipts for agent actions
Curriculum
Work through 5 sections in order.
The chapters are individually addressable documentation pages. You can link directly to a concept from another program, architecture decision, or implementation guide.
Agent-system foundations
Define the agent boundary, assets, actors, flows, and action consequences.
Introduction to governed agents
Model an agent as a bounded decision-and-action system with explicit authority, tools, memory, and evidence.
Define the AI system
Set the intended purpose, actors, assets, dependencies, lifecycle, and system boundary before selecting controls.
Trust boundaries and data flows
Map where trust changes and where data, artifacts, requests, identities, tools, and evidence cross a boundary.
AI assets and attack surfaces
Inventory the data, models, prompts, artifacts, identities, tools, interfaces, infrastructure, and evidence that require protection.
Tools, identity, and authority
Allow only identifiable tools and permissions that are valid for the actor, purpose, and step.
Tool registry, identity, and authorization
Define allowlisted tools, schemas, identities, scopes, credentials, side effects, and revocation behavior.
Identity propagation and policy decisions
Preserve user, workload, service, and tool identity across AI boundaries and make policy decisions explicit.
Permission-aware data access
Enforce source permissions through ingestion, indexing, retrieval, caching, generation, and evidence handling.
Approval and confinement
Bound planning, execution, memory, side effects, budgets, and exceptions.
Approvals, confinement, and memory
Constrain plans and execution with approvals, budgets, isolation, memory boundaries, and deterministic policy checks.
Write effective control objectives
Translate sovereignty goals into testable statements of what must be allowed, prevented, approved, observed, retained, and recovered.
Receipts and assurance
Evaluate behavior, observe operations, retain action evidence, and control change.
Run receipts, evaluation, and incidents
Record attributable action traces and test success, policy compliance, recovery, and incident response.
Evaluation and release gates
Turn acceptance criteria into repeatable promotion decisions with recorded evidence and rollback conditions.
Observability and operational evidence
Design signals that explain service health, AI behavior, policy outcomes, change, and incidents without leaking sensitive content.
Evidence by design
Design evidence alongside controls so important decisions, releases, configurations, and operating events can be verified.
Controlled change and versioning
Version complete AI behavior, assess change impact, approve promotion, support rollback, and retire superseded assets.
Architecture decision records
Record context, options, control consequences, evidence, dependencies, limitations, and reversal triggers for material decisions.
Apply and assess
Produce a governed-agent design and evaluate its control coverage.
Practical completion package
- Agent purpose and authority boundary
- Allowlisted tool and permission registry
- Approval, confinement, and memory policy
- Signed run-receipt schema
- Evaluation and incident-response plan
Current release boundary
SAI-280 remains a roadmap curriculum. The conceptual documentation is public, but executable labs and practical assessment are not released until a narrow tool stack and confinement profile pass independent reproducibility and safety validation.