AI Governance

AI Governance lets regulated and risk-managed organizations put a written AI policy in front of the opportunities Architect generates. You upload a policy document, Linc extracts it into structured constraints, and every automation-type opportunity is evaluated against the active policy. Each evaluation produces an audit-defensible receipt: the constraints that fired, the model's rationale, the policy version in force, and the reviewer's sign-off.

Policies live under Settings → AI Governance; the review queue lives on Oversight → Review queue.

Uploading a policy

1. Go to Settings → AI Governance and click Upload policy for the Initiative you want to govern (PDF only). Initiatives without a policy are unconstrained.
2. Wait for extraction — Linc parses the document into constraints (categories like data classification, PII/PHI handling, human-in-the-loop, approval required, retention, prohibited use), each with a severity: Hard block, Requires review, or Advisory.
3. Review each extracted constraint — confirm or change severity and category, edit wording, dismiss what doesn't apply, approve the rest. Each keeps its source quote and location for traceability. Extraction is a draft, not the final word.
4. Click Activate policy. Once active, it governs new artifacts generated in that Initiative.

Versions: every policy is Draft (doesn't govern anything yet), Active (at most one per Initiative), or Archived (kept for history, still referenced by the receipts it produced). Uploading again creates a new version; activating it archives the previous one. Managing policies is restricted to Initiative admins/owners and org admins.

Organization default and overrides: governance is designed around an organization-default policy with per-Initiative overrides. An override can be Layered (applies on top of the org default — both enforced) or Replace org default (only the override's constraints apply). The bound policy and exact version are recorded on every receipt.

How opportunities are evaluated

Governance evaluates automation-type opportunities only — other types (optimization, standardization, integration, error reduction) are process-improvement ideas, not AI/automation artifacts. Run a check from the Opportunities dashboard; it needs a completed analysis run, an active policy, and at least one automation opportunity. Each evaluated opportunity gets a status:

• Clean — passed all constraints.
• Needs review — one or more constraints flagged it for human review.
• Blocked — a hard-block constraint is violated; proceeding requires a reviewer override with a justification.

When an Initiative has an active policy, the Opportunities dashboard adds a Policy column with a governance badge per automation opportunity (or Unchecked); click a badge to open the receipt.

Activating a new policy version makes traces from the prior version stale — Linc surfaces them so you can re-evaluate under the current policy. Re-evaluation appends a new trace; the original receipt and its sign-off stay in history.

Chat & message governance

When a chat is tied to an Initiative with an active policy, each assistant response is evaluated against the same policy in the background. The message carries a governance shield showing its status, and flagged answers surface in the review queue alongside opportunity traces. Evaluation runs after the response is generated, so governance never delays the chat itself.

The review queue (Oversight → Review queue)

Lists every artifact needing human review. Clean traces don't appear; reviewed traces drop out when a reviewer signs off.

1. Click Open receipt on a row.
2. Work through each constraint evaluation. For flagged or violated constraints, record a disposition: Acknowledged, Accepted risk, Dismissed, or Fixed in artifact — risk acceptance, dismissal, and fixes require a justification an auditor will read.
3. Record the trace-level decision: Approve, Reject, or Override. Blocked traces can't be approved — proceeding requires an override with justification.
4. For stale traces, use Re-evaluate (or Re-evaluate all).

The governance receipt

The receipt drawer is the audit record for one item: the bound policy and version, per-constraint evaluation (status, rationale, required action), reviewer dispositions and justifications, and the final decision with a signature line. Download exports a human-readable JSON copy for audit handoff.