An AI-Orchestrated Architecture for Responding to FOIA Requests

Jason R. Baron, Devanshu Kejriwal

Keywords: Freedom of Information Act, user experience (UX), chatbot, generative AI, technology assisted review, sensitivity review, agentic AI, FOIA exemptions

TL;DR: US FOIA processing is almost entirely human-review centered. This paper proposes an AI-orchestrated architecture with five sub-routines—intake, search, review/redaction, explanation, and audit—each with human-in-the-loop auditing.

Abstract:

In the United States, at every stage of the process for handling requests filed under the Freedom of Information Act (FOIA) and
corresponding state laws, the process almost entirely involves human interaction and review. The goal of replacing the "as is" model of FOIA processing with an AI model is to eliminate obvious bottlenecks due to labor-intensive actions performed by FOIA staff. In this paper we propose a high-level AI-orchestrated architecture for FOIA processing, in which five sub-routines each have their own distinct FOIA task to perform, including: an intake interface, a search protocol, a review protocol for filtering/redacting exempt material, an
explanation protocol for the requester to understand what decisions have been made on withholdings, and software that functions
to audit the process. At each stage, the process anticipates human-in-the-loop review, for the purpose of auditing, quality control,
sampling, and process improvements, and other interventions as needed.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 16