AI document redaction is becoming part of the US government’s plans to process public records requests. The Department of Homeland Security (DHS) plans to use Google’s AI tools to recommend information that should be removed from documents released under the Freedom of Information Act (FOIA). DHS employees will review the recommendations before records are released.
According to documents reviewed by The Washington Post, US Customs and Border Protection (CBP), an agency within DHS, will use Google’s AI tools to recommend which parts of documents may need to be redacted. The planned rollout could cover a significant share of the department’s FOIA workload, raising questions about how AI should be used when government agencies decide what the public can see.
DHS Plans to Apply AI to a Large Share of FOIA Requests
The documents indicate that DHS expects to use AI-assisted redaction on records representing more than 10 percent of its FOIA requests by the end of September 2026. That could involve approximately 100,000 to 140,000 documents, according to the reported figures.
FOIA gives the public a way to request records from federal agencies. However, agencies may withhold certain information under legal exemptions, including material involving personal privacy, national security, or protected government processes.
Reviewing and preparing records for release can be time-consuming, particularly when requests involve large collections of documents. AI tools could help staff identify potentially sensitive passages and reduce some of the manual work involved.
In this case, Google’s technology is expected to make recommendations rather than independently decide what information will be hidden. DHS employees will review the records and the suggested redactions before documents are released.
How AI Document Redaction Works
AI-assisted redaction involves using software to examine documents and flag content that may fall under legal or administrative restrictions. Instead of requiring staff to locate every potentially sensitive passage manually, the system can highlight material for further review.
For the DHS project, Google’s AI tools will recommend information for FOIA officers to black out. Human reviewers will then assess those recommendations and determine what should actually be withheld.
This distinction matters because a recommendation is not the same as a legally justified redaction. FOIA decisions require agencies to apply the relevant exemptions to the records they receive, and automated systems may not understand the full context of a document.
A phrase that appears sensitive in isolation, for example, may be appropriate for public release when considered in context. Conversely, information that looks ordinary could reveal protected details when combined with other material.
Human review is therefore an important part of the planned process.
Why the Government Wants to Automate Document Redaction
Federal agencies handle substantial volumes of public records requests, and processing them can involve searching, reviewing, and preparing extensive files for disclosure.
The DHS documents say automation is expected to significantly reduce processing times. If the tools work as intended, staff could spend less time locating possible redactions and more time reviewing difficult cases.
Faster processing could be useful for both agencies and people waiting for records. However, speed is only one measure of success. Agencies also need to ensure that released documents are accurate, that protected information is handled appropriately, and that information is not withheld without a valid reason.
The DHS project will therefore be watched not only for its ability to reduce workload, but also for how reliably its recommendations support human decision-making.
The Freedom of Information Act (FOIA) gives the public the right to request access to federal agency records, although certain information can legally be withheld. As explained on the official FOIA website, the process helps people obtain government information while recognizing exemptions designed to protect sensitive material. AI-assisted document redaction could help agencies review these records more efficiently, but human oversight remains important to ensure that information is withheld only when the law allows it.
Other US Agencies Are Also Exploring AI for Redaction
The DHS initiative is part of a broader effort across the federal government to explore AI tools for handling records and administrative work.
The Washington Post reported that the Interior Department uses Microsoft tools to identify and redact potential attorney-client information. The Justice Department is also using a tool called aiWARE, developed by Veritone, for video redaction.
Meanwhile, the Department of Health and Human Services is developing a system called FRED to help identify information that may need to be removed from records. A Food and Drug Administration spokesperson told the newspaper that the agency is responsibly exploring limited AI uses to help FOIA staff process records more efficiently.
These examples involve different agencies and types of information. They illustrate how AI may be applied to specific parts of the records process rather than replacing every stage of document review.
The Risks of Using AI to Redact Public Records
AI could make document processing more efficient, but mistakes could affect access to public information.
One concern is over-redaction. If a system flags too much material as sensitive, the public could receive documents with important information unnecessarily hidden. Under-redaction creates a different risk: protected personal or confidential information could be released.
Context is another challenge. Legal exemptions can require careful interpretation, and a system that identifies patterns in text may not consistently understand why a particular passage should or should not be withheld.
There is also the question of accountability. When an AI system recommends a redaction, agencies need clear procedures for reviewing the recommendation, documenting decisions, and correcting errors. The responsibility for releasing records remains with the government agency, not the software provider.
For that reason, the quality of human oversight and the agency’s review procedures will be central to evaluating the project.
What AI Redaction Could Mean for Government Transparency
The use of AI in FOIA processing highlights a broader question about the role of automation in public administration. Tools that help agencies manage large workloads could make government services more responsive, but they also need safeguards when they influence access to public information.
The DHS plan places AI in a supporting role: identifying potential redactions for employees to assess. Whether this approach improves the process will depend on how accurately the system identifies information, how consistently staff review its suggestions, and whether agencies can detect and correct mistakes.
As more federal departments experiment with AI, transparency around how these systems are used will be important. Clear review standards, documented decisions, and ways to challenge improper redactions can help ensure that efficiency does not come at the expense of public access.
For now, the DHS initiative represents a move toward AI-assisted document review, not a transfer of FOIA authority to an automated system. The key test will be whether it can help employees process records faster while preserving the legal protections and public accountability that FOIA is designed to support.
Stay Ahead With Future Tech Hub
Technology moves fast. We keep you ahead with the latest AI, technology, cybersecurity, software, and gadget news.