Government

Make the human decision visible in public-service AI.

Agents can prepare assessments and correspondence. Keep a named official responsible for the governed action that changes a person’s outcome.

A recommendation becomes consequential when a service acts on it.

Define where the system must pause, who has authority to decide, and what an internal reviewer will need to reconstruct the decision later.

Workflow

Issue a proposed eligibility determination.

KLA evaluates the configured rule at the integrated action boundary. A required approval creates a Decision Request in Decision Desk. The downstream action waits for its authorized outcome.

Explore human approval

Decision Desk

Review required
Proposed action
Issue a proposed eligibility determination
Configured rule
Require caseworker approval before issue
Decision owner
Authorized caseworker
Execution state
Hold the notice until a human decision is recorded

Evidence for review

Find the request, the review, and the result.

Case reference, proposed determination, supplied context, applicable policy, reviewer identity, rationale, and issue result.

Lineage Explorer connects recorded events. Evidence Room packages collected evidence into a Sealed Evidence Bundle for verification. Evidence integrity establishes whether the recorded material changed; factual accuracy still requires domain review.

Explore execution lineage

Other workflows to explore

Citizen correspondence

Review consequential agent-drafted notices before sending.

Records disclosure

Route a proposed release or redaction to an authorized reviewer.

Case updates

Apply configured permissions before an agent changes a service record.

Your team determines the applicable obligations and approval rules for each deployment. Explore Control Mapping.

Bring a workflow your team wants to test.

Select one service decision with its operational, technical, and information-governance owners. Confirm the supported route, data handling, review process, and evidence requirements.

Discuss an evaluation
Government AI Runtime Controls | KLA