How AI influences human decisions in DHS surveillance

The U.S. Department of Homeland Security (DHS) has refused to classify several immigration enforcement AI tools as "high-impact AI" on the grounds that "humans ultimately review AI outputs," thereby avoiding transparency obligations required by OMB Memorandum M-25-21, such as AI impact assessments, appeal channels, and public feedback. The article argues this rationale rests on a false assumption: AI outputs, at the moment of generation, have already selected information sources, ignored content, and framed expression; users cannot "unsee" them after viewing, and their subsequent judgments are inevitably influenced by the output. The article cites Anduril's "Autonomous Surveillance Tower" as an example: the system alerts border agents after detecting targets such as people or vehicles, and DHS claims it is "merely a prompt" and thus not high-impact, but the author argues it independently decides what to ignore and flag, not a telescope-like extension of vision.