← UnGovr Oversight

AI Oversight

Concept

Overview

Government agencies are adopting AI for a widening set of tasks — eligibility decisions, risk scoring, fraud detection, chat interfaces, and identity matching. The systems are often procured quietly, and the policies that govern their use are uneven.

AI Oversight is a public record of how government uses AI and how that use is overseen. It documents which systems agencies run, from which vendors, for which decisions, and scores the oversight structures that apply. It reports these facts; it does not argue a position. The project is United States–first.

Use, not law. AI Oversight documents government's use of AI in practice. The law about AI — statutes and regulations — lives separately at law.ungovr.org/ai The two cross-link.

What It Documents

  • Systems in use — decision systems, predictive tools, chat assistants, and identity/face matching.
  • Vendors — who supplies each system, on what contract terms.
  • Decisions affected — where an AI system informs a benefit, a risk score, or an enforcement action.
  • Policies & assessments — whether the agency published a use policy or an algorithmic impact assessment.
  • Disclosure — what is released versus withheld behind vendor trade-secret claims.

Face recognition sits at the seam between AI Oversight and watchwatch — the camera and deployment belong to surveillance, the matching algorithm and its accuracy belong here. The two cross-link rather than duplicate.

The Five Tracks

The same dataset, viewed five ways:

01
Systems & vendors
What AI is deployed and who supplies it.
02
Procurement
Contracts, sole-source awards, and pilots that became permanent.
03
Decisions
Where AI informs eligibility, risk, or enforcement — and due-process safeguards.
04
Policies
Published AI-use policies and algorithmic impact assessments.
05
Disclosure
What the public can obtain versus what is sealed as a trade secret.

The Oversight Score

Each jurisdiction is scored on which oversight structures exist around its use of AI — not on the technology. Where a jurisdiction does not respond, the item is recorded as “not disclosed,” never left blank.

# Example jurisdiction profile (illustrative)
published_use_policy partial
public_assessments no
disclosed_systems yes
governing_oversight partial
score 44 / 100

Built on Public Records

Every data point traces to a public-records request — procurement files, contracts, use policies, and impact assessments. Coverage, recipients, and deadlines are driven by UnGovr's per-jurisdiction records-law analysis, and each request stays within the law in every case. The underlying analysis is published at law.ungovr.org/records/us

Part of UnGovr Oversight

AI Oversight is one member of UnGovr Oversight, a family of observatories that turn opaque corners of government into public datasets with a consistent oversight score. It will live at aioversight.org.

Get Involved

We're looking for journalists, civic technologists, and researchers to help shape what the observatory tracks first. Your experience with government AI procurement and use will inform what we build.

Tech Stack

Public Records Engine Entity Registry Records Law DB Oversight Score Procurement Data PostgreSQL