AI Governance Assessment Checklist

Does this system create new capacity in the people, organizations, and environments it touches — or does it draw them down?

• We can name a concrete capability the system builds in its users over time.
• We have evidence the system does not deskill, fatigue, or displace the people it depends on.
• Environmental and energy costs are measured and accounted for in the business case.
• Value created by the system is shared with the communities whose data or labor enables it.

Is the system aligned with the culture, laws, ecosystems, and values of the people using it — or is it generic?

• Affected stakeholders were consulted before deployment, not only after.
• The system has been evaluated against the specific regulatory regimes that apply in each market.
• Language, accessibility, and cultural assumptions have been reviewed by people from those contexts.
• Edge cases for marginalized or under-represented users have explicit, tested handling.

Do the people affected by this system remain the authors of decisions it shapes — or is it quietly optimizing them out?

• Users can understand, in plain language, what the system is doing on their behalf.
• There is a meaningful, low-friction way to override, contest, or opt out of automated decisions.
• Operators are trained to intervene and have the authority to do so without escalation.
• Defaults are designed so the human-in-the-loop is a real choice point, not a rubber stamp.

Does the system strengthen the relationships between people, institutions, and the wider world — or fragment them?

• The system creates shared context across teams instead of siloing information.
• It surfaces, rather than hides, the people and labor behind its outputs.
• Dependencies on third-party models, data, and infrastructure are mapped and disclosed.
• Failure modes degrade gracefully and preserve human relationships and trust.

Does the system account for its tangible consequences in the physical world — energy, materials, attention, behaviour, wellbeing, and ecological impact?

• We can account for the system's energy, material, and ecological footprint across its lifecycle.
• We have measured the system's impact on attention, behaviour, and wellbeing of people affected by it.
• The system's embodied consequences are documented and reviewed as part of deployment decisions.
• We have criteria for what makes the system's material impact life-giving rather than extractive.

Is GRACE embedded in everyday work through governance, rituals, decision protocols, audits, feedback loops, and learning practices — or does it live only in statements?

• We have named owners, decision protocols, and rituals that keep GRACE alive in daily work.
• Governance is evaluated on a defined cadence with documented criteria, not claimed once at launch.
• Incidents, near-misses, and user complaints feed back into governance decisions.
• Policies, model cards, and data provenance are versioned and publicly traceable where appropriate.