A glossary of key terms and concepts in the AGEI governance framework. Understanding these concepts is essential for implementing evidence-based AI governance.
AGEI is not another AI monitoring dashboard. AGEI is an evidence infrastructure layer that records governed AI activity as cryptographically verifiable receipts, enforces governance gates at runtime, escalates high-risk decisions to human reviewers, and materializes audit-ready bundles on demand.
Think of AGEI as the black box flight recorder for AI systems—but one that records policy decisions before consequential actions occur.
The architectural pattern underlying AGEI. CIAF-LCM separates continuous evidence capture (lightweight receipts) from heavy audit materialization (audit packs). Governed events create compact receipts in real time. Full evidence bundles are materialized later when triggered by audits, legal reviews, or inspections.
This keeps runtime overhead low while preserving the ability to reconstruct governed AI activity when proof is required.
The append-only, tamper-evident storage layer where evidence receipts are recorded. The vault supports tenant isolation, cryptographic verification, and efficient retrieval for audit pack materialization.
A control point in the AI lifecycle where policy rules are enforced. Gates evaluate lifecycle events (model deployment, agent action, data processing) against policy logic and produce outcomes: Approve, Deny, Escalate, Inspect, Require Approval, or Require Elevation.
Gates separate evaluation_status (pass/fail/warning) from gate_outcome (approve/deny/escalate).
The process of applying policy rules to an AI event payload. Policy evaluations produce structured results including evaluation_status (pass/fail/warning), gate_outcome, reason_code, and metadata. Each evaluation creates an evidence receipt.
A materialized bundle of evidence receipts, policy versions, human decisions, and verification metadata. Audit packs are created on demand for audits, legal reviews, or inspections. They include cryptographic proofs of integrity and can be exported as signed JSON, PDF reports, or verification manifests.
A tamper-evident record of a human reviewer's approval or denial decision. Includes reviewer identity, justification, timestamp, and cryptographic proof. Human decision receipts are immutable and auditable.
The process where an agent or system requests additional permissions to perform a high-risk action. Elevation requests go through policy gates and may require human approval. Granted elevations are time-limited and recorded in evidence receipts.
An evidence receipt created when an action is blocked by a policy gate. Includes reason code (policy violation, missing privilege, unauthorized action) and proof that the denial was enforced before the action occurred.
An asynchronous process that validates the cryptographic integrity of evidence receipts. Verification jobs recompute content hashes, check signatures, validate receipt chains, and generate verification manifests for audit packs.