Ethical AI
AI that assists — and never decides the outcome.
Where AI is and is not used, and why it is held to the same checkable standard as everything else Veraq does.
AI is a tool at Veraq, and it is kept in its place. It helps people understand the system, navigate the open record, and read a wave in plain language. It does none of the work that decides an outcome. That work belongs to a published mechanism, commit-reveal-v1, which is committed to the open record before a wave runs and provable after it ends.
The line is firm on purpose. Anything that shapes an outcome must be verifiable by anyone. A model’s internal reasoning is not, so it is held away from the part of the system that has to be checkable. Where AI is used, it is held to the same standard as everything else here: you should be able to see what it touched, and confirm it touched nothing it should not.
What AI does
Explains the record
It reads the open record into plain language, so a participation, outcome or contribution can be understood without reading raw entries.
Helps you navigate
It answers questions about how a wave works and points to the exact entries that support each answer, rather than asking you to take a summary on faith.
Supports moderation and safety
It flags suspected abuse and policy breaches for human review. It surfaces candidates; it does not pass judgement.
Assists the team
It drafts documentation, summarises changes, and speeds internal work. Every output a participant sees has a human accountable for it.
What AI never decides
No model selects a wave outcome, sets a participant’s chances, or alters a defined share. Selection is the commit-reveal mechanism alone, published in advance and reproducible from the record. A model cannot reach that path, and there is no privileged route by which it could.
Nor does AI quietly arbitrate the things that must stay auditable: which cause receives a contribution, how an entry is written, whether a result stands. Those are governed by the mechanism and the record, where they can be inspected end to end. If AI ever influenced an outcome, the outcome would no longer be verifiable, and verifiability is the whole point.
If a thing must be trusted, it must be checkable. AI is used only where it cannot put that at risk.
Ethical AI principles
Assist, never adjudicate
AI informs people; the mechanism and the record decide. The boundary is structural, not a matter of policy we could quietly relax.
Show the working
Where a model makes a claim about the system, it cites the entries behind it, so the claim can be confirmed against the open record.
Bounded and disclosed
Each use of AI has a stated scope and a stated limit. We name what it is for and what it is not trusted to do.
Accountable to a person
A human owns every AI-assisted decision that affects a participant. The model is the assistant; the responsibility does not transfer to it.
Human oversight and provenance
- 01
Scoped before use
Each AI use is defined and bounded before it ships, with an explicit limit on what it may touch and a person accountable for it.
- 02
Held off the deciding path
The selection mechanism runs independently of any model. No AI output feeds the commit-reveal process or the record’s authoritative entries.
- 03
Reviewed by people
Where AI flags abuse or drafts a participant-facing answer, a person reviews before it has effect. AI surfaces; humans decide.
- 04
Recorded so you can check
What is decided about a wave is written to the open record, where its provenance is the mechanism and the entries — not a model you cannot inspect.
Limits and disclosures
Could a model influence which outcome a wave produces?
No. Outcomes come from commit-reveal-v1, committed before the wave and reproducible from the record. AI has no path into selection and no way to alter a result.
Where might AI be wrong?
Its explanations and summaries can err, like any assistant. That is why every claim it makes about the system points to the underlying entries, so you can check the source rather than trust the summary.
Is AI used to set a participant’s chances or change my defined share?
No. A participant’s chances follow from the published mechanism and the defined share is a published, consistent allocation. Neither is set by a model.
How would I know if this changed?
Where AI is used, its scope is disclosed. If a use expanded, the disclosure would say so. The deciding parts of the system stay verifiable on the open record regardless.
AI helps you read the system. It is never what you are asked to trust.