The Department of Homeland Security has posted a notice of intent to award Deloitte Financial Advisory Services a blanket purchase agreement to expand artificial intelligence tools inside SecureRelease, the FOIA case-management software the agency already uses to process public-records requests. The contract would add automated redaction and document-review features on top of software DHS has run for years, according to contracting notices and reporting by FedScoop and other outlets that track the deal.
The volume driving the push is real. DHS started fiscal year 2025 with 221,068 pending Freedom of Information Act requests, took in more than a million new ones over the year, processed nearly that many, and still ended the year with a backlog of 245,572. DHS receives more FOIA requests than any other federal department, spanning ICE, Customs and Border Protection, the Transportation Security Administration and the Secret Service.
SecureRelease’s AI features work by pattern-matching: the system comes loaded with a template of commonly redacted categories, then “learns” from the manual edits human reviewers make, applying those patterns to future documents. It is not the first AI tool in the department. CBP’s FOIA division already runs RelativityOne, a platform that learns from past reviewer decisions and can independently tag whole documents as relevant or irrelevant to a request, sometimes without a human flagging them first.
That expansion is happening as the human side of the operation shrinks. More than 600 federal information specialists, the staff who actually review FOIA requests, have left government service since the start of the second Trump administration, according to transparency researchers tracking attrition across agencies. Fewer people are available to check what the software decides, at the same time the software is taking on more of the decision.
Transparency advocates say that combination is the problem. “AI trained on existing responses will reflect that,” Abigail Kunkler of the Electronic Privacy Information Center said of the approach. “Using AI to automate redactions almost guarantees that DHS will over-redact.” Cody Venzke of the ACLU put it more bluntly: “FOIA is more than just a keyword search. Redactions are a legal question with significant nuance. AI is simply not ready for those sort of decisions.” Patrick Eddington of the Cato Institute has separately warned that letting software independently code documents as irrelevant risks letting sensitive, high-profile requests skip the layers of review built to catch them.
DHS is not acting alone. The Justice Department’s Office of Information Policy ran a NextGen 3.0 FOIA Tech Showcase for federal agencies this year, and nearly 85 percent of the vendors who showed up were marketing AI tools for exactly this kind of work: flagging relevant records, drafting redactions, transcribing and masking audio, running public-facing chatbots for FOIA offices. The vendor market assumes agencies government-wide are moving the same direction DHS is moving now.
FOIA exists because the public has almost no other legal claim on what an enforcement agency decides to keep secret. The exemptions themselves, for law-enforcement sensitivity, for deliberative process, for privacy, are already written broadly enough that DHS gets wide latitude to withhold. Training a redaction tool on the agency’s own history of withholding does not introduce a neutral check on that discretion. It automates the discretion, at scale, inside the one department facing the most sustained public demand right now to explain what its enforcement arms are actually doing.
Deloitte’s incentive here is straightforward: SecureRelease is a product the firm sells across the federal government, and a high-profile DHS deployment makes it easier to sell elsewhere. DHS’s incentive is a backlog it cannot process with the staff it has left. Neither of those pressures point toward more disclosure. They point toward a faster machine for making records disappear before anyone outside the agency sees what was in them.