{"slug": "the-influence-of-ai-on-human-decisions-in-dhs-surveillance", "title": "The Influence of AI on Human Decisions in DHS Surveillance", "summary": "The Department of Homeland Security has declined to classify its AI-driven immigration enforcement tools as high-impact, exempting them from the disclosure, impact-assessment, and appeal safeguards required under the Trump-Vance administration's April 2025 OMB Memorandum M-25-21, according to research analyst Emily Froude of Democracy Forward. DHS justifies the exemption by claiming a human always reviews AI output before action, a defense Froude argues rests on the fallacy that humans can independently evaluate AI output without being influenced by it. Froude cites CBP use of Flock automated license plate readers, an ICE app for mapping \"potential deportation targets,\" and DHS face-recognition apps used on protesters and observers as examples of AI surveillance that warrant scrutiny.", "body_md": "# The Influence of AI on Human Decisions in DHS Surveillance\n\nEmily Froude / Sep 21, 2026\n*Emily Froude is a research analyst at Democracy Forward. The views expressed in this piece are the personal perspectives of the author.*\n\nThe stories trickle out in sputters and gasps. United States Customs and Border Protection (CBP) agents [using](https://www.404media.co/email/1925e57b-2f02-4180-babb-2302f64638aa/?ref=daily-stories-newsletter) Flock automated license plate readers that can track cars across the entire nation. An app that Immigration and Customs Enforcement (ICE) agents use to map “[potential deportation targets](https://www.404media.co/elite-the-palantir-app-ice-uses-to-find-neighborhoods-to-raid/?ref=daily-stories-newsletter)” in a particular area. Agents taking photos of protestors and observers without their permission, using an app the Department of Homeland Security (DHS) says can “[determine or verify](https://www.wired.com/story/cbp-ice-dhs-mobile-fortify-face-recognition-verify-identity/)” their identity. Drip, drip, drip.\n\nGovernment agencies [are deploying](https://www.nextgov.com/artificial-intelligence/2026/04/agencies-report-over-3000-ai-use-cases-2025/412898/) more artificial intelligence applications than ever before. AI used by DHS deserves [particular scrutiny](https://www.techpolicy.press/dhs-ai-surveillance-arsenal-grows-as-agency-defies-courts), given the agency’s expressed objective to uproot immigrants from their communities, tear them away from their families, and deport them to countries where they could face persecution or even death. The administration’s [own AI policy](https://www.whitehouse.gov/wp-content/uploads/2025/02/M-25-21-Accelerating-Federal-Use-of-AI-through-Innovation-Governance-and-Public-Trust.pdf#page=3) calls for strong disclosure standards for AI systems that drive decisions about people’s rights and safety. But DHS has declined to classify its AI-driven immigration enforcement tools as high impact, and so it avoids these requirements altogether. Its justification is that a human always reviews the AI’s output before acting. But this defense rests on a fallacy: the myth that humans can independently review and act on an AI output without being influenced by that same AI output. To hold DHS accountable for how it is actually using these AI tools in immigration enforcement, we must confront the truth about how AI impacts human decision-making.\n\nIn April 2025, to comply with the [AI in Government Act of 2020](https://www.congress.gov/116/plaws/publ260/PLAW-116publ260.pdf), the Trump-Vance administration’s Office of Management and Budget (OMB) issued a government-wide AI policy, [Memorandum M-25-21](https://perma.cc/BT92-ZLVY), “Accelerating Federal Use of AI through Innovation, Governance, and Public Trust.” The memo largely carried over protections first put in place under the Biden-Harris administration’s own OMB AI policy — a rare point of continuity between the two administrations, reflecting a shared recognition that AI used by the government has the potential to significantly harm the public’s rights and safety. The policy directs agencies to meet a set of minimum safeguards for AI uses that “could have significant impacts when deployed,” which it calls “high-impact AI.” While agency chief AI officers can waive any of these requirements, these safeguards include completing AI impact assessments to document the potential impacts of the AI system on people’s privacy, civil rights, and civil liberties; ensuring adequate human training; giving individuals affected by AI-enabled decisions the opportunity to appeal; and giving the public an option to submit feedback on the AI use case.\n\nThe Trump-Vance OMB memo defines high-impact AI narrowly: a system whose “output serves as *a principal basis* for decisions or actions that have a legal, material, binding, or significant effect on rights or safety” (emphasis added). The “principal basis” language in this definition enables agencies to argue that, as long as a human technically makes the final call, the AI’s output was never the principal basis for anything — no matter how much that output shaped the decision. DHS has relied on this loophole to worm out of classifying some of its most alarming uses of AI as high-impact and avoid the associated transparency requirements. In its most recent [AI Use Case Inventory](https://www.dhs.gov/ai/use-case-inventory), DHS claimed that several AI tools that were “presumed to be high impact” under the Trump-Vance OMB memo because of their ability to adversely affect rights and safety were *not* actually high impact because the agency did not fully rely upon the AI outputs in making decisions or taking subsequent actions.\n\nAt the root of the issue with DHS’s classification system is that AI outputs do more than provide information; they exercise independent decision-making that humans using the information they produce cannot always see. By the time an output is generated, the AI system has selected what sources to draw from, what information to ignore, and how to frame its response. That an AI system makes judgments means it is more than simply a source of information that can be independently verified by users. Additionally, as with anything a user reads, an AI system’s outputs cannot be unseen. Just as a human’s memory [might be colored](https://www.cambridge.org/core/journals/memory-mind-and-media/article/fake-memories-a-metaanalysis-on-the-effect-of-fake-news-on-the-creation-of-false-memories-and-false-beliefs/B6A66A305AFD6724E1C1C56A68BB5CF2) by misinformation consumed and not verified, what a user learns from an AI system after they prompt it sticks with them, especially if they cannot meaningfully double-check the system. The output becomes part of the basis for what the user does next. That means that the actions they take after viewing the system’s output are necessarily impacted by the system itself, even when they believe they are exercising independent judgment. This is precisely the flawed assumption that DHS relies on to keep its AI tools off the list of those subject to heightened disclosure.\n\nTake the [explanation](https://www.dhs.gov/sites/default/files/2026-07/26_0612_mgmt_cx_2025-june-update-dhs-ai-use-case-inventory.xlsx) for DHS’s “Autonomous Surveillance Tower” AI system. According to the use case inventory, this Anduril-made system “provides alerts when it detects the presence of an [item of interest] (i.e., persons, vehicles, animals) in the image frame.” The system then determines whether the item is likely to be a person and sends an alert to border agents. DHS claims that this system is not high impact because “the AI merely alerts to the presence of an item it was trained to detect.” But the system is not just serving as an extension of the border agents’ sight, like a telescope or a pair of binoculars; it independently selects what to ignore and what to flag. Unable to scan the entire border themselves, border officials are entirely dependent on the system’s output: the Autonomous Surveillance Tower’s alerts dictate where armed government forces are dispatched. The system’s classification, in other words, is the *principal basis* for DHS’s deployment decisions. Any arrest that follows will have a legal, material, or binding effect, and even a mere encounter with DHS agents can have a significant effect on a migrant’s rights and safety. By the administration’s own definition, the Autonomous Surveillance Tower should be a high-impact use case.\n\nAnother example is one of DHS’s most notorious AI apps known to date, Enhanced Lead Identification and Targeting (ELITE). ICE [uses](https://www.404media.co/elite-the-palantir-app-ice-uses-to-find-neighborhoods-to-raid/?ref=daily-stories-newsletter) this Palantir app to view a map of potential deportation targets, with each address scored for the likelihood the target lives there, alongside a list allegedly prioritizing so-called high-value targets — possibly immigrants with a criminal history or an outstanding deportation order. DHS [claims](https://dhs-ai-explorer.onrender.com/use-case/DHS-2578) that ELITE is not a high-impact AI use case because its “outputs are limited to normalized addresses” and officers “review and validate the AI-driven outputs before determining actions.” But, in reality, ELITE is selecting the targets. ICE officers are not reviewing, nor likely even familiar with, the algorithm assigning confidence scores or generating prioritization lists. It is not possible for them to fully review and validate ELITE’s outputs. Instead, agents are likely demonstrating automation bias — an over-reliance on algorithmic output without being “[positioned to exercise meaningful scrutiny.](https://www.techpolicy.press/ai-efficiency-can-undermine-accountability-even-with-humans-in-the-loop/)” ELITE’s entire purpose is to identify and prioritize addresses for immigration enforcement, leading agents to particular locations that they would not otherwise know to target. The system’s outputs are once again the principal basis determining which addresses ICE targets for immigration enforcement. Anyone witnessing DHS appearing at immigrants’ homes in the past 18 months knows that is an action with “legal, material, binding, or significant effects on individuals.”\n\nTo be sure, DHS [is not](https://fedscoop.com/opm-memo-federal-hiring-ai-usage/) the only government agency exploiting this loophole to evade classifying its AI systems as high impact. But the stakes are especially high for DHS. In the agency’s telling, ELITE and the Autonomous Surveillance Tower’s outputs are just one of many factors—but never the principal basis—for deciding which immigrants to target. The claim would be laughable if the consequences were not so heartbreaking.\n\nThe evasion is also unnecessary. Acknowledging the truth about how AI impacts decision-making would not force DHS to stop using these tools. It would merely require DHS to reckon with the limits of human review of outputs, classify its AI use cases correctly, and follow the rules that apply to them. The high-impact designation is not especially burdensome: it triggers an impact assessment, training for officers who use the system, appeal rights for the people it affects, and public consultation. These are basic processes designed to make the government’s conduct more transparent and accountable.\n\nOne might reasonably advocate for going further and ending some of these DHS AI use cases altogether. At a minimum, though, we should expect the Trump-Vance administration to follow its own policy and abide by existing safeguards for using AI. Doing so would allow public oversight of these critical decisions, allowing us to see how DHS is surveilling immigrant and border communities and on what basis. It would wrench open the tap of information that has so far only come out in drips.\n\n## Authors\n\n[Emily Froude](https://www.techpolicy.press/author/emily-froude)\n\n## Topics\n\n## Related\n\n[The AI State is a Surveillance State](https://www.techpolicy.press/the-ai-state-is-a-surveillance-state)March 12, 2025", "url": "https://wpnews.pro/news/the-influence-of-ai-on-human-decisions-in-dhs-surveillance", "canonical_source": "https://www.techpolicy.press/the-influence-of-ai-on-human-decisions-in-dhs-surveillance/", "published_at": "2026-09-21 16:58:52+00:00", "updated_at": "2026-09-21 17:24:53.962399+00:00", "lang": "en", "topics": ["ai-policy", "ai-safety", "ai-ethics", "computer-vision", "artificial-intelligence"], "entities": ["Department of Homeland Security", "U.S. Customs and Border Protection", "Immigration and Customs Enforcement", "Office of Management and Budget", "Democracy Forward", "Emily Froude", "Flock", "Palantir"], "alternates": {"html": "https://wpnews.pro/news/the-influence-of-ai-on-human-decisions-in-dhs-surveillance", "markdown": "https://wpnews.pro/news/the-influence-of-ai-on-human-decisions-in-dhs-surveillance.md", "text": "https://wpnews.pro/news/the-influence-of-ai-on-human-decisions-in-dhs-surveillance.txt", "jsonld": "https://wpnews.pro/news/the-influence-of-ai-on-human-decisions-in-dhs-surveillance.jsonld"}}