Loading story…
A Lawfare report said the Department of Homeland Security and its immigration agencies have deployed a broad set of AI systems for enforcement and adjudication with limited public visibility and uneven governance. Citing FOIA records, the report said ICE operated about 42 internal AI tools as of May 2024 while only 23 appeared in DHS’s public AI Use Case Inventory, and that CBP operated roughly 75 AI systems, 62 of them active. The systems identified include ICE’s ImmigrationOS and Hurricane Score, as well as USCIS’s Asylum Text Analytics, which are used to prioritize enforcement targets, estimate absconding risk, and flag asylum applications for additional scrutiny.
The report said required safeguards such as privacy impact assessments, testing, and meaningful human review were often delayed or inadequate, including USCIS’s national rollout of Asylum Text Analytics before approval of a required Privacy Impact Assessment. It also described ICE contracts for AI-assisted skip tracing and plans for large-scale social media surveillance, warning that the technology enables enforcement at a scale that can outpace human review and raises civil-liberties concerns. DHS’s public AI inventory provides an official baseline for disclosed systems, but the gap between that inventory and FOIA-derived records intensified concerns over transparency, oversight, and the expansion of executive AI powers in immigration enforcement.
See the reporting duties and controls this puts on the clock.
7 events from the most recent confirmed update back to the earliest known activity.
A December 2025 executive order sought to limit state AI governance through litigation threats and fiscal incentives, according to the article. It describes this as occurring despite the absence of a comprehensive federal AI framework.
In March 2025, ICE agents arrested Columbia University student and pro-Palestinian activist Mahmoud Khalil after DHS claimed his student visa and then his green card had been revoked. The article cites the arrest as an example of immigration enforcement power.
As of May 2024, FOIA-produced records showed ICE operated about 42 internal AI tools, while only 23 appeared in DHS's public AI inventory at that time. The same records showed CBP operated roughly 75 AI systems, including 62 active systems.
DHS published its AI Use Case Inventory Library, which the article later uses as the public baseline for comparing disclosed ICE and CBP AI systems. The reference is a DHS publication page for that inventory.
The Office of Management and Budget's 2024 AI guidance required safety- or rights-impacting AI systems to undergo real-world testing and an AI impact assessment before operational deployment. The article cites this as a benchmark for evaluating DHS immigration AI governance.
FOIA records showed USCIS expanded its Asylum Text Analytics tool nationally before the Privacy Impact Assessment required by the E-Government Act had been approved. The tool is used to flag asylum applications for additional scrutiny and monitor attorneys and preparers as a network.
The article states that post-9/11 authorities such as the PATRIOT Act and the Homeland Security Act helped build enduring U.S. surveillance infrastructure beyond their original counterterrorism purpose. It says these systems were first deployed heavily against noncitizen communities and later expanded more broadly.
See what this changes for your reporting obligations and which controls it puts on the clock.
2 references tracked. Mallory keeps watching after this page renders.
Map indicators from this story to your assets and identify affected systems in minutes.
Every observed campaign, victim, and pivot linked to actors named in this story.
Malware, exploits, and IOCs connected to the activity described here.
YARA, Sigma, and Snort rules deployed to your SIEM as soon as they’re published.
Get matching new stories delivered to your team as they break — not the next morning.
Ask questions about this story and take action on the answers.
Other stories Mallory is tracking that overlap on entities, sources, or topic labels.
The US Department of Homeland Security (DHS) published its 2025 AI use case inventory, showing **law enforcement operations** as the agency’s most common application of AI. Of **238** AI use cases listed across DHS, **86** were categorized as law enforcement, with **Customs and Border Protection (CBP)** and **Immigration and Customs Enforcement (ICE)** accounting for a majority of DHS’s reported AI deployments (CBP **83**, ICE **51**). DHS also reported that **35** law-enforcement AI use cases were designated **“high impact”** under Office of Management and Budget (OMB) criteria, meaning outputs can materially affect rights or safety; examples cited include anomaly detection, smartphone forensics triage, open-source/social media analysis, and biometrics-driven capabilities such as facial recognition. Within the inventory, ICE disclosed an **“AI Enhanced ICE Tip Processing”** capability used to sort, translate, and summarize public tip submissions to help investigators prioritize urgent cases and reduce manual review. Reporting indicates the service uses **Palantir** generative AI tooling to produce a “**BLUF**” (bottom-line-up-front) summary using one or more **commercially available large language models**; DHS stated the models were trained on public-domain data by providers and were not additionally trained on agency data, though they do process tip submissions during operation. The inventory lists the tool as becoming operational on **May 2, 2025**, and notes it was not present in the prior year’s inventory.
Jun 29, 2026
US lawmakers and civil-liberties advocates are escalating scrutiny of **ICE** and **CBP** use of facial recognition amid reports that immigration agents have used face-scanning tools on people observing or protesting enforcement activity. Proposed legislation dubbed the **“ICE Out of Our Faces Act”** would seek to bar ICE/CBP facial recognition use, alongside related demands to limit tracking of First Amendment activity and questions about whether ICE maintains a “domestic terrorists” database tied to immigration protests. Separately, records reviewed by *WIRED* indicate DHS’s *Mobile Fortify* facial-recognition app—rolled out in 2025 to “determine or verify” identities during operations—cannot reliably **verify** identity and is intended at best to generate investigative leads; the reporting also describes DHS approving the tool after weakening centralized privacy reviews and removing department-wide limits on facial recognition. In parallel, Congress is debating the renewal of **FISA Section 702**, with lawmakers raising concerns that domestic surveillance authorities could be applied more broadly, including in support of immigration-related enforcement framed as a national security issue. Nextgov reports that transparency data shows intelligence agencies increased 702 searches in 2024 using identifiers linked to known or suspected Americans while pursuing foreign cyber and terrorism threats, even as the FBI reduced its own direct U.S.-person queries after new safeguards. A separate Nextgov item on an AI moratorium and sector-specific AI regulation is policy-focused and not directly tied to the ICE/CBP facial recognition deployments or Section 702 reauthorization debate described in the other reporting.
Jun 29, 2026
Reporting indicates the U.S. Department of Homeland Security (DHS) has significantly increased its use of **administrative subpoenas**—which do not require judicial approval—to obtain identifying information for anonymous social media accounts that criticize **ICE** or post information about ICE agent locations. According to the New York Times (as cited by TechCrunch), DHS sent **hundreds** of subpoenas to major platforms including **Google, Meta, Reddit, and Discord**, and some companies reportedly complied in at least some cases; Google said it notifies users when possible and challenges subpoenas it views as **overbroad**. Separately, a court filing disclosed that the **IRS improperly overshared** immigrants’ personal data with DHS/ICE under an April 2025 IRS–DHS data-sharing arrangement intended to support certain non-tax criminal investigations under **IRC Section 6103** exceptions. After ICE requested **1.28 million** addresses, the IRS could verify **47,289** individuals, and for a subset (reported as **under 5%** of verified matches) the IRS mistakenly provided ICE with **additional address information** beyond what ICE had supplied, raising concerns that the interagency data-sharing deal increased exposure of sensitive taxpayer information.
Jun 29, 2026