DHS Turns to Google AI for FOIA Redactions, Raising Transparency Concerns

The US Department of Homeland Security (DHS) is reportedly acquiring Google AI and machine-learning tools to assist in redacting sensitive information from Freedom of Information Act (FOIA) records. According to reporting by The Washington Post and Engadget, the technology is expected to apply to more than 10% of FOIA requests, covering an estimated 100,000 to 140,000 submissions.

The DHS aims to ensure that requesters receive accurate and complete information, as stated by a spokesperson. However, critics warn that AI-assisted redaction could become a powerful tool for secrecy if agencies rely on it to reinforce overly broad withholding decisions.

Other federal agencies are also using AI and machine-learning tools for redaction purposes. The Department of Justice (DOJ) uses Veritone software to assist with redacting video, while the Department of Health and Human Services is developing an AI tool called FRED with an unnamed contractor.

While human employees will still review proposed redactions before a final determination, concerns have been raised about the potential for automation bias and overcorrection. Critics argue that agencies need to publish error rates and maintain auditable logs of accepted or rejected AI recommendations to ensure transparency and accountability.