Post-vote, not voter-facing
AI Story is an optional layer after a released result. It does not guide voter choice, choose an option or replace deterministic decision evidence.
AI governance
How Votegrain governs AI
Use AI to help explain released decision evidence. People review its output; it never steers a vote or changes the result.
Include AI in your pilot evaluation, with a limited scope agreed before activation. See the pilot scope.
How much AI is included?
Agree the number of initiatives and the generation allowance in your AI-enabled Order and AI Use Schedule.
Start with released decision evidence. Comparisons need multiple compatible decisions; a higher initiative count alone does not switch AI on.
The boundary
Useful after the vote. Bounded everywhere.
Votegrain keeps AI away from voter choice and inside a controlled server-side path. Each layer can be disabled without removing the underlying decision evidence.
Before publication, an author-only format check can flag structure. It cannot evaluate options, recommend a choice or block publication.
Server-side and allow-listed
Model calls run server-side from allow-listed inputs with strict output validation. Prompt-injection text and user-supplied links are treated as untrusted data.
Feature-specific controls and a stop switch
Each AI feature needs permission and policy checks, with a platform stop switch. Your agreed allowance sets the usage boundary; the underlying evidence remains available without AI.
Human release gates
People stay in the release path.
Generation creates a draft, not a published conclusion. The review and release steps remain explicit, authorised actions.
Drafts stay private
A generated draft is kept private until a person reviews it. The review decision and note are recorded rather than implied by generation.
Release is a separate gate
Review and internal release are separate authorised actions. Source changes, version conflicts or a required second reviewer block release instead of forcing it through.
Viewing is not authority
Viewing Interpretation, Statistical trends or Evidence does not grant AI Story generation, review or release capability. Deterministic reporting remains available while an AI draft is unavailable or under review.
When AI is unavailable
The decision record remains complete.
A model can be unavailable, rejected or withdrawn without changing the released result or inventing a replacement.
Deterministic evidence stays authoritative
If model output is unavailable or validation fails, the failure is reported honestly and the released result is left untouched.
The input is bounded
The governed How we decide AI Story uses deterministic aggregates only, never ballots or identities. Validators reject recommendations, predictions, unsupported causal claims and claims about culture or personality.
The vote remains human
AI never guides voter choice. It is an advisory, post-vote layer for permitted management audiences, not a voting influence or a substitute for a final decision.
Inspect the boundary
Trust is easier to assess when the limits are visible.
Read the wider security model, or bring a governance question to a Votegrain conversation.