Democracy Forward: DHS AI classifications skirt OMB safeguards

"The stories trickle out in sputters and gasps," Emily Froude writes in an essay for Tech Policy Press about the AI tools now running through Department of Homeland Security enforcement work. Froude, a research analyst at Democracy Forward, argues DHS has read a key OMB definition narrowly enough to keep most of these systems out of the "high-impact" tier that would trigger public safeguards.

The pivot is OMB Memorandum M-25-21, issued in April 2025. It defines high-impact AI as systems whose "output serves as a principal basis for decisions or actions" with a legal or material effect. DHS's position, in Froude's telling, is that human review breaks that chain: if an agent looks at an alert before acting, the software is not the "principal basis," and requirements like impact assessments, training, appeal rights and public consultation fall away.

The essay names the tools. An Anduril Autonomous Surveillance Tower, which DHS describes as a system that "merely alerts to the presence of an item it was trained to detect." Mobile Fortify, an app DHS says can "determine or verify" a person's identity from a photograph, including of protestors. A Palantir application ICE uses to map "potential deportation targets" in a particular area. Flock automated license plate readers that CBP agents use to track cars across the country.

Froude's counter is automation bias, "an over-reliance on algorithmic output," and the observation that once an agent sees a tool's flag they are rarely "positioned to exercise meaningful scrutiny" of it. The remedy she urges is procedural: reclassify the tools as high-impact and run them through the safeguards the memo already requires.

Two researchers we track shared the piece the day it went up.