Civic AI is on the receiving end of an accountability conversation that enterprise AI never had to have. When a permit gets denied or a 911 call gets prioritized, the public deserves to understand why. The good news is the engineering is straightforward. The hard work is institutional.
The transparency stack
Innfini for cities ships with a transparency stack that runs alongside every agent. Three layers:
- Decision registers — every algorithmic action recorded with its inputs, outputs, and reasoning.
- Equity dashboards — outcomes broken down by district, demographic, and time. Public-facing.
- Appeal workflows — residents can challenge any algorithmic decision through a structured intake.
What gets published
Every decision that affects a resident's services, permits, or access is logged with: the algorithm version, the input features, the output, the confidence, and an explanation in plain language. The log is queryable by anyone via a public API.
Why this matters
Black-box civic AI is a political liability waiting to happen. Cities that ship transparent decisions don't get caught flat-footed when the press starts asking questions.
The hard part: equity audits
Decision registers are easy. Equity audits are hard. The question "is this AI fair?" doesn't have one answer — it has dozens, and different ones matter to different communities. We work with each city's civil-rights office to define which fairness metrics get tracked and surfaced.
The platform doesn't pick fairness for you. It instruments it. The choice — and the conversation — stays with the public.