US federal agencies are moving to incorporate artificial intelligence into the process of redacting sensitive information from documents requested under the Freedom of Information Act (FOIA). According to GitHub documents reviewed by The Washington Post, the Department of Homeland Security (DHS) intends to utilize Google’s AI tools to identify and recommend data that should be blacked out.
The initiative is part of a broader effort by US Customs and Border Protection to streamline its handling of public records requests. Officials project that by the end of September, AI-assisted redactions will cover more than 10 percent of all FOIA submissions, amounting to approximately 100,000 to 140,000 documents. Agency representatives stated that the automation is anticipated to significantly shorten processing times.
Under the proposed workflow, DHS employees will review the AI’s suggestions to ensure that requesters receive information deemed “accurate and complete.” This terminology has drawn scrutiny from legal experts concerned about potential over-redaction.
The decision follows heightened public interest in government transparency, particularly after the Justice Department released millions of documents related to Jeffrey Epstein with extensive redactions. The current push to automate these procedures comes as other agencies also integrate similar technologies. The Interior Department is employing Microsoft tools to flag attorney-client privilege material, while the DOJ uses Veriton’s aiWARE platform to redact video content. Additionally, the Department of Health and Human Services is developing a proprietary system called FRED to obscure unidentified sensitive data.
Despite the efficiency gains, privacy advocates warn of the implications for public access. A FOIA expert at the University of Florida told The Washington Post that the technology could become a powerful instrument for secrecy, urging careful oversight of how governments utilize AI in document redaction.
Efficiency is great, but ‘accurate and complete’ isn’t cutting it when AI might be over-redacting sensitive but releasable info.
Wait, Google is helping DHS decide what the public sees? Sounds like a conflict of interest waiting to happen.
Black box redactions are a transparency nightmare. How can we trust what stays hidden if we can’t audit the algorithm?