AI-Assisted Audio/Video Redaction

Overview

Development stage

DeployedSource: c) Deployed – The use case is being actively authorized or utilized to support the functions or mission of an agency.

High-impact designation

Presumed high-impact, determined notSource: b) Presumed high-impact, but determined not high impact

Impact justification

The AI-Assisted Audio/Video Redaction use case does not meet the definition of a High-Impact category under OMB M-25-21 due to its narrowly defined scope, reliance on human oversight, and non-decision-making role. This tool is designed to assist Homeland Security Investigations (HSI) by partially automating the redaction of audio and video evidence, such as detecting and obscuring faces, objects, or sensitive information (e.g., license plates, PII), to protect individuals’ identities during investigations and legal proceedings. Importantly, the tool does not perform 1:1 facial matching or identification, and its outputs are strictly limited to redactions, which are subject to comprehensive human review and editing before finalization. This ensures that the AI’s role is supportive rather than determinative, with no direct impact on investigative decisions or legal outcomes.

Start date

2024-07-01 (day precision)Source: 2024-07-01T00:00:00

Withheld from public reporting

Not reported

Mission

Topic area

Law Enforcement

Operational functions

FOIA and records management Derived by keyword rules; see methodology

Problem the AI is intended to solve

This use case intends to solve the problem of the labor-intensive process of redacting audio and video evidence.

Expected benefits

The AI-Assisted Audio/Video Redaction tool is used to reduce the manual effort required to redact audio and video evidence used during an investigation and subsequent legal proceedings.

System outputs

The AI output in this use case are redactions to the media file. Users will further edit the redacted media prior to exporting the final redacted file to ensure completeness. Homeland Security Investigations conducts a human review of each frame within redacted files prior to distribution.

Technology

AI classification

Computer visionSource: Computer Vision: AI that processes and interprets visual data (e.g., images and videos).

System name(s)

Not reported

Custom-developed code

No

Public source code

Not reported

Sourcing

How it was built

Purchased from vendorSource: a) Purchased from a vendor

Vendor (as reported)

Case Guard

Vendors (standardized)

None on the standard list

A vendor is the supplier named by the agency. It does not identify the underlying model or AI technology.

Data and privacy

Involves PII

Yes

Privacy Impact Assessment

www.dhs.gov ↗

Authorization to Operate

No

Demographic features

Not reported

Training and evaluation data

All training sets used for the model are from public and private collections of images. AI models are trained using a combination of real-world and synthetic datasets collected from publicly available sources. These datasets are curated to represent a broad range of conditions, including edge cases such as occlusions and poor lighting, to improve detection accuracy across varied scenarios.

Federal Data Catalog

Not reported

Governance

0 of 8 minimum-practice questions answered. A blank answer means the agency reported nothing; it does not mean the practice is absent.

Potential impacts and how they were identified

Not reported