Legal — draft
Responsible AI
A draft statement of the commitments Guardian intends to hold itself to. Placeholder language pending review.
DRAFT PLACEHOLDER — this document is illustrative only. It has not been reviewed by legal counsel, creates no obligations or guarantees, and will be replaced before public launch.
1. Separation of authority
Guardian is designed so that what it may know and what it may do are distinct, with action authority always narrower and always human-granted.
2. Knowing when not to act
Where stakes are high, permission is absent, or the situation is ambiguous, the intended behaviour is to stop and ask rather than proceed.
3. Uncertainty disclosure
Guardian aims to state what it does not know rather than presenting fluent language as confidence.
4. Human review and correction
People can question, correct or reverse what Guardian surfaces, and review a record of actions taken on their behalf.
5. Learning limits
Learning from corrections is intended to improve judgment only. It should never widen the scope of what Guardian is permitted to do.
6. Sensitive contexts
Health, financial, legal, children's and belief-related contexts receive heightened restraint. Guardian is not a substitute for professional judgment.
7. Conscience protection
Within Common Ground, Guardian commits to accurate representation of worldviews, no covert persuasion, and leaving conscience with the person.
8. Independent review
Placeholder: external assessment, audit and reporting practices will be described here once established. No certification is claimed today.
Questions about this draft can be sent to the Guardian team. Final documents will be published following counsel review.