The ICE–Lexis nexus: An argument against use of commercial databases in immigration enforcement

Abstract

In 2022, immigration advocates filed a lawsuit against commercial data broker LexisNexis (“Lexis”) for its role in facilitating unwarranted surveillance of immigrants and political activists critical of U.S. immigration policy (Castellanos et al. v. LexisNexis Risk Solutions, Complaint, 2022). The plaintiffs allege that use of Accurint (the product Lexis sells to U.S. Immigration and Customs Enforcement) violates consent, compensation, and privacy laws, further subjecting noncitizens to monitoring that can result in severe enforcement action—including detainment, deportation, and arrest. This lawsuit reflects a broader phenomenon in which brokers partner with law enforcement and intelligence agencies to collect, aggregate, and analyze personal data, build information systems to make inferences about individuals, and sell their services to aid policing activities. In this article, we argue that U.S. government use of Accurint is morally objectionable. The Castellanos complaint outlines harms connected to freedoms of speech, association, equal rights, and consumer protection. Lexis’ response points to government interests in enforcing immigration laws, public safety, and conducting criminal investigations. We argue that government use of sociotechnical decision systems such as Accurint are justifiable only to the extent that such use adequately respects the autonomy and agency of persons subject to those systems, which turns on reliability, purported responsibility, stakes, and relative burdens. Finally, because federal use of Accurint conflicts with the legitimate policy goals of substate actors, its use for enforcement of noncriminal immigration violations is unjustifiable.

Introduction

A stereotypical image of U.S. immigration and customs enforcement involves a range of visible surveillance activities: agents patrolling the U.S.–Mexico border, questioning individuals at ports-of-entry, monitoring CCTV in detention centers, and inspecting vehicles and shipping containers. Less visible are agencies’ ¹ intelligence capabilities, including their increasing reliance on commercial data brokers (e.g., LexisNexis (“Lexis”), Thomson Reuters, Palantir) to aid in immigration enforcement. These brokers have long-term, close working relationships with law enforcement and intelligence agencies, in which they collect, aggregate, and analyze personal data, build information systems to make inferences, and sell their services to aid policing (Lamdan, 2022; Villa-Nicholas, 2023; Wang et al., 2022).

Immigration and Customs Enforcement’s (ICE) intelligence practices affect a wide range of people, including those seeking to enter the United States, those advocating on behalf of immigrants, and journalists reporting on immigration and law enforcement. This article analyzes one prominent example: Lexis’ nexus with ICE.

In 2022, Lexis was sued for its role in facilitating unwarranted surveillance of immigrants and political activists (Castellanos et al. v. LexisNexis Risk Solutions (henceforth “Castellanos”), Complaint, 2022). The plaintiffs allege that use of Accurint (the product Lexis sells to ICE) violates consent, compensation, and privacy laws, further subjecting noncitizens to enhanced monitoring that can result in severe enforcement action—including detainment, deportation, and arrest. The legal harms address a “heightened risk” for misidentification, the denial of opportunities, and a lack of control over sensitive personal information (cf. Funk, 2019; Moreno, 2017). The plaintiffs also voice First Amendment concerns, alleging that such expansive targeting discourages advocacy work critical of the state. ² Ultimately, the case was dismissed for a lack of plaintiff standing. ³

At first glance, Castellanos is a straightforward claim about the legal rights of individual persons and an organization subject to state surveillance in the context of immigration enforcement. Setting aside the legal merits, there is plausibly a moral wrong here. Our goal is to better understand whether, and the conditions under which, ICE's use of Accurint is morally objectionable.

We argue that U.S. government use of Accurint is morally objectionable. The complaint outlines harms connected to freedoms of speech, association, equal rights, and consumer protection. Lexis’ response points to government interests in enforcing immigration laws, public safety, and conducting criminal investigations. The values that Castellanos adduces are foundational to a liberal democratic society, and how to ensure those values while carrying out the government's legal obligations is an evergreen and perennially thorny question.

We argue that government use of sociotechnical decision systems like Accurint are justifiable only to the extent that such use adequately respects the autonomy of persons subject to those systems. That requires that persons be able to reasonably endorse the use and to exercise appropriate agency (Rubel et al., 2021). It also requires that government entities be able to fulfill policy goals within their legitimate purview. The “reasonable endorsement” test we describe turns on reliability, purported responsibility, stakes, and relative burdens to persons subject to Accurint. The argument is grounded in persons’ abilities to exercise reasonable control over the course of their lives, evaluate where they stand vis-à-vis instruments of state power, and dispatch their responsibilities within a democratic state. The government legitimacy test turns on discrete functions of state and substate authority.

Based on the limited information available about Accurint, our best assessment is that no one subject to the system could reasonably endorse it. Hence, it is unjustifiable. Finally, because federal use of Accurint conflicts with the legitimate policy goals of substate actors, its use for enforcement of noncriminal immigration violations is unjustifiable. These arguments are distinct from arguments about the justifiability of immigration laws and enforcement generally (see, e.g., Costa, 2021; Lister, 2020; Sager, 2017). However, the moral bases of those arguments help ground the reasonable endorsement and agency arguments.

Background: Castellanos v. LexisNexis

In Castellanos, the plaintiffs assert several legal harms related to the collection and sale of their “sensitive personal data.” The complaint states that Lexis generates revenue by collecting and selling U.S. consumer data without consent, allegedly in violation of Illinois privacy and consumer protections. Lexis sells these services to corporations, law enforcement, and government agencies via its online platform, Accurint.

Accurint is a database that compiles public and nonpublic information. It relies on the large-scale collection and aggregation of consumer transaction data, allowing users to make connections across disparate data points to create encyclopedic profiles of individuals. Lexis uses the data collected to generate identity profiles of millions of U.S. consumers. ⁴ Accurint is marketed to law enforcement agencies. It uses law enforcement data not publicly accessible, such as real-time booking information, to determine a person's past and present location. Accurint also gathers information that would otherwise require a subpoena or court order for law enforcement to access (e.g., utility records). ⁵

The plaintiffs allege several harms. Some are political activists critical of policing and immigration policy. They state that Lexis deprives them of the ability to control their sensitive information, which increases their risk of being targeted for their political advocacy (either by law enforcement or others). There is also a potential harm given the sensitivity of information compiled: current addresses, cell phone numbers, social security numbers and private information about persons’ relatives, neighbors, and associates. The plaintiffs allege that this data compilation heightens their risk for “identity theft, stalking and other potential injuries” (Castellanos et al., 2022, para. 6).

Other plaintiffs are immigration advocacy organizations. ⁶ They note additional harms, such as fear of retaliation for political speech, arrest, and deportation of members or people in their network, misidentification, and being denied opportunities in regular consumer transactions. For example, arrest and deportation are possible due to ICE's ability to pull real-time location data from Accurint. The denial of opportunities and misidentification are possible due to potential inaccuracies.

The complaint alleges a range of causes of action related to a lack of consent and compensation in data collection, sharing, and use. It states that these data practices deprive persons of the economic value of their personal data. The plaintiffs assert that they have “suffered invasions of privacy,” which resulted in mental distress and a wider fear among people in their network (Castellanos et al., 2022, para. 73). That fear is based on potential exposure to a range of harms resulting from the links made by ICE's use of Accurint.

Lexis filed a motion to dismiss. The Castellanos plaintiffs responded, and various pretrial motions were litigated. As of April 2024, Lexis’ motion to dismiss was granted for a lack of plaintiff standing.

Critiques of data systems and immigration practices

Understanding surveillance in society

There are numerous potential critiques of ICE and Lexis. One is to situate the Lexis case against the surveillance studies literature, which generally examines surveillance through its structure and processes. ⁷ This literature is too large to synopsize here. However, there are some key threads that speak to surveillance technologies similar to Accurint, to surveillance in the context of immigration and borders, or to the economic structures that catalyze development of Accurint and similar systems. In this section, we consider several of these as a way of placing and differentiating our arguments.

This literature is often based on architectural perspectives of surveillance, understood as centralized and spatial, resembling Bentham's panopticon. ⁸ More contemporary theories see surveillance as decentralized and digital—signaling a “postpanoptic” turn—where control hinges on access to distributed and networked data. Haggerty and Ericson's (2001) surveillant assemblage captures this shift to describe the convergence of once discrete surveillance systems into a reassembly of abstracted, virtual “data doubles” for classification. Extensions of this include Gandy's (2021) panoptic sort, where corporations rely on surveillance to categorize consumers by their “presumed economic or political value,” and Lyon's (2002) social sorting concept, which describe the discriminatory effects of everyday surveillance (see also Brayne, 2022). Surveillance, in this sense, relies on digital infrastructures to classify and manage populations through personal data.

Surveillance at the intersection of capital and state

The case could also be examined through another distinct line of critiques. These focus on the connection between data, surveillance, and capital.

One line engages directly with political economic theories of information. For example, Gandy (1993) describes how difficult it is to define information, let alone quantify its value. He argues that classical and neoclassical economic theories fail to capture information's distinct characteristics ⁹ during late-stage capitalism (Gandy, 1993). Another example, in a postpanoptic take, is Zuboff's (2019) account of surveillance capitalism—a “new economic order” marked by pervasive collection and profit from personal data. ¹⁰ This economic logic removes the need for markets to cater toward “the genuine needs of populations and societies,” thereby representing a threat to democracy and persons’ capacity for self-determination (Galič et al., 2017 referencing Zuboff).

A second set of inquiries tackle the links between state and corporate players in the coproduction of the surveillance society: the “surveillance–industrial complex” (Ball and Snider, 2013). The economic, political, and social dynamics between corporate and government agencies enable a billion-dollar market for surveillance tools (Hayes, 2012). These markets converge with government agendas. A variation of this phenomenon emerges along national borders: the “border–industrial complex.” For example, Molnar (2021) describes how borders have many layers of surveillance. These layers create a powerful migration control sociotechnological system through which state, corporate, and international (nonstate) organizations operate, and often engage in technological experimentation with limited oversight. ¹¹

The market through which Lexis operates illustrates a manifestation of these accounts. A small number of third-party data brokerages amount to what Lamdan (2022) calls “data cartels.” These companies (such as Lexis’ parent company RELX and Thomson Reuters) engage in surveillance capitalism, and operate surreptitiously across a range of industries to commodify our data. ¹² Packaged in the form of “risk assessment” tools, data brokers sell our data to government entities and other institutions. ¹³

Technology, surveillance, and ethics

The surveillance critiques are impressive in scope. Many scholars approach surveillance by understanding its core structures and processes, rather than judging it as inherently good or bad, as Marx (2015) suggests. While this is an important task, it is not the focus here.

The architectural and infrastructural perspectives offer abstracted ways of thinking about surveillance through its structure. Gandy and Lyon tackle the social structures that organize surveillance practices. Zuboff's target is an entire economic system: capitalism as it shifts toward surveillance and data collection, aggregation, commodification, and use. Lamdan's critique is an entire industry that forms a key part of Zuboff's economic structure. The data cartels she examines take the data exhaust that Zuboff describes and turn it into products used across many industries. The surveillance–industrial complex interrogates these synergies at the intersection of private and state. Underlying each of these critiques is an understanding that something has gone wrong, and that surveillance capitalism (per Zuboff) or data cartels (per Lamdan) are bad.

Our task is to argue for why at least one facet of this phenomenon is unjustifiable. If surveillance capitalism is somehow wrong, if the panoptic sort is bad, if the surveillance–industrial complex is malign, and if data cartels are unjustifiable, they are so because of the wrongs that constitute them. Analyzing the moral justifiability of using Accurint in immigration enforcement helps explain why certain deployments of surveillance are morally problematic (if they are). Moreover, we are in the midst of surveillance capitalism right now, we have been panoptically sorted, and we walk among data cartels. That may change, but it is difficult to see surveillance capitalism dissipating tout court or data cartels withering on the vine. If those do happen, they will happen because facets of surveillance capitalism and particular data cartel practices are curtailed by determining what conditions make use of products like Accurint in immigration enforcement unjustifiable, and why.

It is worth clarifying what we mean by “justifiable.” ¹⁴ Our basic premise is that actions can be morally better or worse, which is to say they can be more or less morally justified. ¹⁵ This is meant to be a relatively minimal commitment, for several reasons. First, we do not assume that any particular normative moral theory is uniquely valuable; that is, we do not assume that all moral value is reducible to, say, consequences, virtue, duty, or egalitarianism. Rather, we recognize that a range of important moral values will bear on any particular question. Second, the view recognizes that conclusions are revisable. It is difficult to pin down everything that matters morally and to weigh different values against each other, and it is difficult to ensure that all relevant facts about cases are addressed thoroughly.

More specifically, our argument is premised on a set of moral values, grounded in agency and autonomy, which provide a foundation for the moral claims we explain below. While agency and autonomy are foundational to our argument, we consider them in light of other values, and do not claim that those are the only morally considerable values.

Immigration ethics

There is a broad literature surrounding immigration enforcement and ethics. This ranges from questions about immigration law and doctrine, to moral questions about immigrant rights and state powers, to empirical questions about enforcement effectiveness and the relationship between migration and other goods (e.g., economic growth, wages, employment). In the “Iimmigration, Accurint, and reasonable endorsement” section, we draw on arguments about proportionality in law and immigration enforcement, the potential for immigration enforcement to be arbitrary and (hence) contrary to principles of republican freedom, and about the different roles of state (i.e., national) and substate (e.g., state and local) governments in the context of immigration.

Another thread concerns ethical matters beyond particular immigration laws, such as whether (and under what conditions, if any) immigration restrictions are morally justified. Some scholars argue that national borders themselves are unjustifiable, and that disadvantaging people not born in well-resourced states is unjust (Carens, 1987). Others stop short of that view, but still maintain that U.S. immigration restrictions are now (and recently have been) unjustifiably restrictive. For this article, we assume some restrictions are morally permissible, and states have limited powers to enforce restrictions, although there are certainly moral limits on what those restrictions can be.

It is also worth noting the distinction between moral claims made about immigration enforcement generally versus claims against the use of information technologies. For this project, we do not set out to make a general case against immigration enforcement, but rather aim to incorporate features from the immigration ethics literature (some of which make moral claims against enforcement) to make a case against ICE's use of Accurint. Beyond immigration, these technologies raise their own set of moral questions that must be considered separately, which we turn to next.

Immigration, Accurint, and reasonable endorsement

Decision systems and reasonable endorsement: Basics

One question we have not addressed is the justifiable use of technological tools. All of the issues raised in Castellanos are plausibly problematic. ¹⁶ What is important is that those concerns are linked to the use of technologies themselves. Concerns about the use of big-data and predictive technologies are general. They are relevant across many domains, though they interact with questions specific to immigration enforcement. The immigration context allows us to examine moral questions surrounding the justifiable use of such technologies.

Reasonable endorsement

Rubel, Castro, and Pham (RCP) examine several data-driven, algorithmic decision systems, including criminal justice risk-assessment systems and teacher evaluation systems (Rubel et al., 2020). While each of these technologies has a degree of intuitive promise, each has flaws. Determining when such systems are justifiable is difficult to pin down. RCP argue that a justification for algorithmically-aided decision tools requires understanding how they relate to autonomy, agency, and respect for persons. They argue that respecting individuals subject to automated-decision systems requires considering whether those systems can be reasonably endorsed by persons subject to them.

In short, persons are autonomous and capable of determining the values that will structure their lives, to the extent they see fit. Moreover, respecting a person's autonomy “requires recognizing them as value-determiners, neither thwarting nor circumventing their ability to act according to those values without good reason” (Rubel et al., 2020). It also requires recognizing that people are capable of abiding fair terms of social agreement (so long as others do too). These “moral powers” set the parameters for how people ought to treat each other. Good reasons for circumventing persons’ ability to act according to their values are “those reasons that they can abide as fair terms of cooperation” (Rubel et al., 2020) citing (Rawls, 1999; Scanlon, 1998). Policies and practices are endorsable to the extent that they either align with persons’ values or are fair terms of social cooperation. ¹⁷

Here it is worth explaining endorsability. First, our argument is that endorsability is a necessary condition for justifiability. It is not the criterion for what actions, policies, laws, or states of affairs are best. There will be a range of policies and practices that are endorsable, but bad for other reasons. Second, for an action (law, policy, etc.) to be reasonably endorsable by someone is not the same as that person in fact endorsing the action (cf. Scanlon). People have ranges of views and situations that cannot be easily abstracted. For example, individuals with high incomes may not endorse progressively graduated income taxes. They would prefer (and might only be willing to endorse) flatter taxes, since flatter taxes decrease their individual taxes and would advance their individual goals. However, it is plausible that people with high incomes could (hypothetically) endorse progressively graduated income taxes because it would comport with fair terms of social cooperation (supporting cooperative governance and social services). The framework allows us to consider circumstances from others’ perspectives even while recognizing that individuals have highly varied views.

Translating this general view about justification and reasonable endorsement into prescriptions about use of information systems is a further question. RCP argue that there are four questions in determining whether such a system may be reasonably endorsed by persons subject to them: (1) whether the system is reliable, (2) whether the system bases inferences on actions for which subjects are responsible, (3) the stakes of decisions based on outputs of such systems, and (4) the relative burdens of such decisions on subjects.

Consider RCP's example of teacher evaluation systems, such as EVAAS ¹⁸ , used to determine how effectively teachers help students learn. Such systems may be more or less reliable in measuring student learning and connecting it to teachers; the more reliable, the more plausible it is that teachers could endorse such systems. But reliability is not enough. A system might have reliable measures but those measures might be a function of things for which a teacher is not responsible. It might reflect student social disadvantage or facts about teachers that they have no responsibility to change. For example, an evaluation system might reliably show that teachers are less effective in a school where persistent poverty affects learning. It is difficult to see how teachers could endorse such a system. Even reliability and responsibility are not the full picture; the stakes involved also matter. If teachers used evaluation systems to assess their own practices, then even marginally reliable systems might be endorsable. This changes if the stakes are higher, such as promotion, pay, or job loss. High stakes decisions are not by themselves unendorsable; rather, that turns on how reliable the system is and whether it turns on issues for which the subjects are responsible. Finally, there is a question of relative burden. Decision systems do not work in isolation, and an important issue is whether they create additional burdens for people or groups that are subject to other pressures.

We apply this framework to ICE's use of Accurint.

Reliability

The first facet of reasonable endorsement is reliability, meaning a system in some sense provides good information. Lexis claims that Accurint is accurate, boasting a 99% accuracy rate. But reliability is more complicated. It involves a combination of error rate, purported use, and actual use. Put another way, reliability in one context does not confer reliability in adjacent contexts. To explain, let us canvas what the purported and actual uses of Accurint are.

Lexis describes Accurint as an online database used by government, law enforcement, and commercial entities to help assess and manage risk. ¹⁹ Lexis markets Accurint as a “locate-and-research tool” that uses “proprietary data-linking technology” to access its vast repository of public and nonpublic information to quickly identify individuals (Welcome to Accurint, n.d.). Beyond identification, Accurint can link individuals with relatives, specific locations (e.g., address), employers, and any property or assets a subject might own (businesses, vehicles, etc.). According to Lexis, Accurints’ value lies in its ability to identify people even when the data is sparse, fragmented, outdated, or missing.

Lexis frames Accurint as a tool for managing risk “intelligently.” The company helps its clients prevent fraud, enhance security and protect citizens, increase efficiency and effectiveness, and minimize financial loss. Lexis points to a variety of use cases and clients in law, healthcare, insurance, and banking. ²⁰

Lexis provides services to government and law enforcement agencies. The company's identification and linking features aid in investigations (Accurint^® for Government, n.d.c). Lexis advertises Accurint as a tool for law enforcement agencies to easily access “hard-to-find” information, connect data across local, regional, and national systems, identify patterns of criminal activity, and broaden an agency's jurisdictional view (Accurint^® Crime Analysis, n.d.b; LexisNexis^® Accurint^® Virtual Crime Center, n.d.).

Lexis contracts with ICE, but says their services promote public safety and national security. It is also used in the deportation context, but claims this use is limited to serious threats “including child trafficking, drug-smuggling and other serious criminal activity” (LexisNexis Risk Solutions, n.d.a). ²¹

While there is no official statement by the DHS or ICE about its use of Accurint, journalists describe the agency's broad application of the tool. Intercept reporter Sam Biddle surfaced an internal email from ICE's Enforcement and Removal Operations (ERO), confirming Accurint's use in deportation. The email states that Accurint should be “widely utilized by ERO personnel” for the “identification, location, arrest, and removal of noncitizens” (Biddle, 2022). There is also evidence of ICE using Accurint's crime database to conduct regular searches and reports (Biddle, 2022). This includes daily queries by ICE's division of Homeland Security Investigations (HSI), which has faced criticism for its broad national security directives and reported involvement in targeting immigrants with nonserious criminal backgrounds.

Lexis appears to recognize the problem of using Accurint in different contexts. It highlights the circumstances under which it entered its current contract with the DHS and ICE in March 2021, stating that it believed the agencies would focus on migrants with “serious criminal backgrounds.” This appears to distance Lexis from some of the most egregious ICE policies from 2016 to 2020, including family separation and harsh detainment conditions. Lexis states that

Under the Biden Administration, in March 2021 LexisNexis Risk Solutions was awarded a contract to provide an investigative tool to the Department of Homeland Security's U.S. Immigration and Customs Enforcement. We entered into this contract understanding that the mission of Immigration and Customs Enforcement under the Administration had changed to focus immigration enforcement resources on people with serious criminal backgrounds. (LexisNexis Risk Solutions, n.d.a; emphasis added)

Nothing in this statement precludes different emphases over time.

Accurint may be reliable across its actual use cases, but their purported use and purported accuracy may not extend to those uses. Moreover, the difference in public justification and actual uses creates a problem for the legitimacy of the more aggressive forms of surveillance used against migrants.

Even taking as true Lexis’ claims about accuracy and assuming it holds across contexts, there remains a concern related to misidentification. While Accurint claims a 99% accuracy rate, it also acknowledges that the “data sources used in the reports may contain errors” and it is “sometimes reported or entered inaccurately, processed poorly or incorrectly, and is generally not free from defect” (LexisNexis^® Accurint^® Virtual Crime Center, n.d.). The inaccuracies among data brokers who compile information from public records and commercially available data are also well reported (Moreno, 2017). A report commissioned by Mijente reveals that Lexis “furnished a million-dollar insurance policy alongside the contract for potential ‘errors and omissions’ resulting in lawsuits” with the City of Denver (Mijente et al., n.d.). As impressive as Lexis’ figures are, there remain grounds for concern about inaccuracy. We discuss the importance of stakes related to such inaccuracies below.

Finally, there is an issue regarding reliability over time, where the goals of using a system like Accurint change. Accurint plausibly does a good job of predicting violations of immigration law. But which violations of immigration law gain the attention of enforcement agencies varies dramatically over time, by presidential administration, by public mood, and more. This means that the purported justifications of immigration enforcement will vary as well, and Accurint's success on our reliability criteria will similarly vary.

Here it is worth noting that reliability is just one criterion for justifiable use of a predictive system. Even if we assume that Accurint is reliable—that is its predictions are overwhelmingly accurate—reasonable endorsement of Accurint in the immigration context will turn on several other factors.

Responsibility

In the criminal justice and teacher cases, RCP argue that a key issue in whether persons can reasonably endorse a system is the degree to which that system affects them based on actions for which they are responsible (Rubel et al., 2021). Use of a risk assessment system (such as the well-known COMPAS tool) to inform sentencing that predicts likelihood of reoffense based on addiction, housing insecurity, and family status would index punishment to conditions for which a defendant is not responsible. ²² Such uses (at least when the stakes are high) are ones persons could not reasonably endorse, regardless of whether they reliably measure student academic achievement (in the teacher cases) or predict recidivism (in the COMPAS case).

Applying this conception to immigration enforcement has multiple parts. First, there are innumerable reasons why people move or overstay visas in violation of immigration laws. Among these are fleeing violence, seeking economic opportunity, connecting with family, responding to incentives of employers, and more. In many cases, those who violate immigration laws may not be morally responsible for their actions. One may have an obligation to provide a safe environment or seek economic advantage for their family and it is probably a moral imperative to maintain family and social ties. One cannot be morally liable—which is to say culpably responsible—for undertaking an action for which one has an all-things-considered moral obligation to perform.

Here we can draw on elements of the immigration ethics literature. For example, Brock (2020) argues that migrants who have resided in the United States for a long time have claims against enforcement action. Receiving states may be obligated to provide pathways for “regularizing status.” These claims are based on “long-settled migrant's” relationships with the state, community, and family. ²³ The passage of time is also important. How long a migrant can expect to reside in the United States creates conditions where migrants must develop situated plans, language skills, and other competencies to meet basic needs (economic, social, etc.). ²⁴ Disrupting these plans can result in severe harms, including unnecessary hardship or distress. ²⁵

Following this line of reasoning, there are many ways that actions for which people are not responsible affect violations of immigration law. In Brock's argument, some long-settled migrants have a claim against deportation (asylum-seekers, children of migrants, or temporary status holders). When enforcement falls short of the state's commitments to respecting the rights of migrants (e.g., avoiding worsening people's conditions), migrants have reasonable claims to stay in the receiving state, even if it means evading the law. Similarly, Carens (1987) argues that immigration restrictions unjustifiably privilege those living in well-resourced states, and that birthplace should not dictate a person's ability to migrate.

In both Brock's and Carens’ views, immigration restrictions are unjustifiable when they make people worse off based on luck of birthplace (Carens) or harm people acting in response to circumstance (Brock). Certain kinds of immigration restrictions are themselves morally wrong, and people do not have a moral obligation to follow such restrictions. While the use of immigration enforcement tools draws on facts for which people are causally responsible (i.e., they undertook some action) and legally responsible (i.e., immigration law prohibits their actions), people are in many cases not morally responsible. And that is the relevant sense of “responsible” at work here.

The question of responsibility in information systems goes beyond whether people are morally responsible for migration. The more troubling issue is that Accurint draws on the data trails from immigrants’ engagement with key components of civic responsibility. The ways people, including migrants, engage with society and act as responsible community members leave substantial data trails: paying taxes, purchasing utilities, applying for driver's licenses, reporting criminal conduct, or enrolling in school (Lamdan, 2022). A risk assessment system can be legitimately used contrary to persons’ interests. However, a system based on actions for which people are not morally culpable (often the case in migration) and where the effect is to penalize persons providing for their basic needs or engaging with their communities in prosocial ways, is not a system that persons can reasonably endorse.

Here we want to be careful not to overstate our argument. Brock's argument concerns cases where immigration enforcement is itself unjustified. If Brock is right, then using technology like Accurint to enforce immigration law is (a fortiori) unjustifiable. But we need not reach the question of whether Brock is correct in all such cases to draw on her argument. The argument is based on the kinds of activities and relationships that migrants have, and the fact that those kinds of activities are key elements of social responsibility. Our argument is that basing enforcement actions on exercises of responsibility (via information collection about those activities) is something that makes enforcement less justifiable, independently of whether enforcement can in some cases be justified.

Returning to Castellanos, one through-line is the connection between use of essential services and information within Accurint. The plaintiffs point to how ICE can use Accurint to search cell phone, water, and electricity records. They write “[b]ecause of these powerful tools, noncitizens who share their information while purchasing essential utilities or engaging in other common consumer transactions risk ICE enforcement action” (Castellanos et al., 2022, para. 5). They allege that Lexis includes a broad range of information to build its Accurint tool (e.g., utilities, cell phone, Internet, cable TV, insurance, employment, rental, licensing (driver's, vehicle), and credit check). Lexis maps family, social, community, and other associational networks as well (Castellanos et al., 2022, paras. 23, 24, and 50).

It would be unreasonable to ask immigrants to refrain from using essential services. Further, law-abiding immigrants are encouraged to integrate into society. But the more they do, the greater the risk of being identified by Accurint and of deportation. ²⁶ Not only is Accurint a system that keys on facts for which persons are not responsible, it conflicts with the very foundations of civic responsibility. Migrants could not reasonably endorse use of Accurint as a matter of fair terms of social cooperation precisely because it conflicts with the underpinnings of such cooperation.

Stakes

Systems might be problematic because they are unreliable or address actions for which people are not responsible. But such a system could be endorsable if the stakes are low. For example, use of algorithmic decision systems tracking teacher performance ²⁷ might be reasonably endorsable if used for self-evaluation purposes. They would not be reasonably endorsable where the stakes are higher (contract renewals, promotions, bonuses).

Here, the stakes for undocumented immigrants are high. There are myriad reasons why immigrants would not endorse ICE's use of Accurint. For one, there is the possibility of being detained and deported. This is the most alarming concern, which presents significant harms. ²⁸

Also at stake are family and social ties. The consequences can be life-changing, especially for immigrants who are long-time residents, or who have American-born children and family members. Immigrants also establish community ties and interdependent social networks. These social ties can be crucial to individuals’ well-being and to the welfare of other community members.

There are also substantial economic stakes in immigration enforcement. Individual deportation and detention (including the threat of it) disrupts employment and labor networks, leading to financial harm and decreased opportunities. The Castellanos plaintiffs also argue that people subject to Accurint's system are being deprived of the economic value of their personal data; those are lower stakes than deportation and job loss, but important nonetheless.

Fear of misidentification is another troubling component. While Lexis claims an error rate of less than 1%, the company acknowledges misidentification as a possibility. Not all cases of misidentification have significant consequences (they may be easily resolved). However, the stakes of misidentification can be high, and can lead to severe harms, including arrest, detainment, and deportation. ²⁹ The potential for misidentification extends to family and friends, who are subject to these systems by association.

Lastly, threats to fundamental, democratic liberties are at stake among both citizens and noncitizens, including the right to freedom of association and political speech. ICE's use of Accurint has a “chilling effect” among journalists, political activists, and others critical of U.S. immigration policy (Jones et al., 2019). ³⁰

There is another related point. Numerous criticisms of algorithmic systems concern transparency and explainability. The ability to understand the workings of such systems is important for preserving a person's practical agency (i.e., their ability to control aspects of their lives) and cognitive agency (i.e., their ability to understand themselves and their relation to social contexts). ³¹ While the practical impact of transparency within an immigration context may be limited due to other powerful forces and currents affecting the lives of migrants, it is nonetheless important because the stakes are high.

Relative burden

The last principle is relative burden, which asks: do these systems impose unfair burdens to those subject to their use? Algorithmic decision-making systems must account for a set of tradeoffs, therefore a distribution of benefits and burdens is inevitable. Yet, the extent to which these tradeoffs matter and impose a discriminatory social effect depends on several factors. The burdens imposed must neither (a) be arbitrary, (b) reflect morally unjustifiable distinctions, nor (c) compound existing burdens to members of socially disadvantaged groups (Rubel et al., 2021). With respect to ICE's use of Accurint, we evaluate each of these factors.

The first considers whether a burden is arbitrary. A burden is arbitrary where it has little to do with a system's purported use and where that burden is unpredictable. To understand this sense of arbitrariness, consider the stop-and-frisk policies that resulted in a series of lawsuits with New York City's police (Floyd v. City of N.Y., 959 F. Supp. 2d 540 (SDNY 2013)). These policies permitted officers to stop, interrogate, and search residents based on suspicion of criminal conduct, falling short of the “probable cause” requirement for search under the U.S. Fourth Amendment. ³² Yet, despite the scale and frequency of stop-and-frisks, only a small percentage of stops resulted in arrest. More alarming was the disproportionate impact stop-and-frisk had on Black and Hispanic residents, who comprised a majority of stops. This raised concerns over the policy's effectiveness in identifying criminal activity or weapons possession. The frequency, scale, and possible inefficacy of this practice, which was divorced from its justification, created an environment where NYC residents were persistently exposed to arbitrary power. ³³ Costa argues that immigration enforcement has similar hallmarks of arbitrariness and hence violates neo-republican freedom. ³⁴ Others extend these critiques of arbitrariness and discretion toward the bureaucratic practice of immigration administration. ³⁵

Use of Accurint resembles NYC stop-and-frisk. Its use is decoupled from rationale (as stop-and-frisk was decoupled from Terry v. Ohio (1968) rationale). Its use is pervasive, out of the blue, and makes people persistently subject to exercise of arbitrary power (as with NYC stop-and-frisk). To justify working with ICE, Lexis claims Accurint is an effective tool for identifying serious criminal violations. Yet evidence suggests that ICE's broad and routine use of Accurint to target undocumented migrants contradicts this rationale, raising plausible concerns around abuses of power (Biddle, 2022).

Advocates raise alarm regarding ICE's use of Lexis to circumvent sanctuary city ordinances that aim to protect immigrant communities (Mijente et al., n.d.). This concern is compounded by ICE's dubious track record of targeting migrants with no criminal history or records with minor offenses (TRAC, 2019). HSI regularly uses Accurint; but HSI is often criticized for involvements in workplace raids, family separations, and other invasive practices targeting immigrant communities (Biddle, 2022). The arbitrary exercise of power and use of decision systems are also subject to fluctuations in presidential administration, where discretion and enforcement vary widely in response to evolving priorities. From 2017 to 2020, ICE arrests of people without criminal convictions and long-term residents increased dramatically (Funk, 2019).

In early 2025, there has been a marked increase in immigration enforcement across federal agencies: revocation of visas; detention for speech- and politically related activities; arrest of a federal immigration judge; active avoidance of due process; deportation of Venezuelan nationals to a notorious El Salvadoran prison (NYC Bar Association, 2025). These actions are ongoing at the time of this writing, and fully incorporating those into our analysis would be premature. However, there is reason to suspect that the actions increase the salience of arbitrariness and proportionality.

A second consideration for fairness behind a system's distribution of burdens is whether those burdens reflect morally unjustifiable distinctions. This idea is akin to the well-developed literature on proportionality in enforcement of a mala in se criminal law (Walen, 2023). Intensity of enforcement of and penalty for criminal law should reflect the seriousness of offense. That same idea applies in the context of mala prohibita immigration restrictions. In criminal law, moreover, there is a powerful argument for statutes of limitations to prevent punishment for old offenses. ³⁶ Brock (2020) argues that a similar kind of proportionality holds for enforcing immigration laws against people who have lived in a place for a long time.

Turning back to surveillance technologies, Accurint expands ICE's power to enforce immigration restrictions substantially. While ICE may aim to prioritize the pursuit of individuals with serious criminal convictions, in practice this pursuit is contingent on persons who are findable (Funk, 2019). On-the-ground enforcement turns on how easy it is to pursue specific individuals rather than on whether those individuals violated serious criminal laws. Enforcement based on ease (aided by Accurint) is at least prima facie unjustifiable because it flips proportionality on its head. That is, ease of enforcement is correlated with the degree to which potential targets are participating in social and civic life, are law-abiding, and engaged in prosocial actions (Lamdan, 2022). The data trails people leave behind, and those easiest to find, are from such individuals. ³⁷ Therefore, to the extent that behaving responsibly in society correlates with enforcement in a way that targets these individuals (as opposed to people who are not engaged) is an unjustifiable difference.

The last consideration is whether a system's burden compounds existing burdens to members of socially disadvantaged groups. Despite the unavailability of information about the decision tradeoffs calculated and measured within Accurint, we know that those who are subject to ICE's enforcement paradigm are from vulnerable groups. ICE's use of Accurint is compounding an already discriminatory system that privileges some and penalizes others. The effects of these systems compound the existing burdens on communities of color who are more likely to be targeted and arrested by ICE in the first instance.

Levels of government

So far, we have applied the Reasonable Endorsement Test to ICE's use of Accurint, and outlined its four key elements: reliability, responsibility, stakes, and relative burden. Each speaks to either a subjects’ rationality (i.e., having some end that structures subjects’ actions) or reasonableness (i.e., being a fair term of social cooperation that subjects can abide so long as others do too). The responsibility element contained two parts. The first regarded the degree to which Accurint reflects facts for which subjects are responsible. The second concerned the ways Accurint turns elements of civic and social responsibility on their head, creating liabilities for people engaging in basic civic and social responsibility.

There is a further argument related to the responsibilities of state and local government. In the U.S. constitutional context, federal, state, and local municipalities structure the hierarchy of governance. Such arrangements, like in many other national contexts, establish norms of subsidiarity. ³⁸ In this case, state and local governments encourage migrants to engage in basic acts of civic and social responsibility by insulating community members from federal enforcement with so-called “sanctuary laws,” which restrict cooperation with ICE in some deportation efforts. This is an operationalized form of the second of the two moral powers described above, namely abiding fair terms of social cooperation so long as others do too. It is important for society that migrants seek employment, pay taxes, enroll in school, and communicate with police and social services. But where such participation (i.e., abiding fair terms of social cooperation) results in high stakes sanctions like deportation (i.e., contrary to “others do too”), anyone would be reluctant to do so.

The immigration ethics literature raises a related argument, and asks who should enforce restrictions. Lister (2020) argues that it is crucial to consider the roles of government actors at different levels, and that immigration enforcement must respect the responsibilities of substate actors, rather than (or in addition to) the moral claims of migrants. State-level actors (federal actors in the United States) have a different set of obligations than substate entities (individual states and municipalities). He argues that substate actors may justifiably “refuse to comply with laws” that would prevent them from fulfilling essential functions (Lister, 2020). Substate actors are responsible for providing law enforcement, public health, and education. When federal agencies enlist state and local governments to aid in immigration enforcement, this impedes on substates’ ability to fulfill their responsibilities, conflicting with sanctuary laws. ³⁹ In 2021, Cook County, Illinois rejected over 1000 detainer requests from ICE, demonstrating a county-level policy against deportation despite ICE policy to pursue removal. ⁴⁰ Tools like Accurint allow ICE to obtain real-time incarceration data and circumvent local government decisions to release people into their communities.

Lister's argument is similar to the reasoning that underwrites state and local sanctuary policies. Substate governments have grounds to refuse to be enlisted into enforcement of federal immigration laws, and Accurint provides a workaround. By using Accurint as a workaround to enlist local governments in immigration enforcement, ICE obviates the right to refuse to be enlisted. Use of Accurint undercuts state and local governments’ ability to carry out their responsibilities in the domains of their legitimate power. The reason substate actors have a right to refuse to enforce federal immigration policy is that enforcing that policy makes migrants less able to act as responsible community members. ⁴¹ Enforcement via Accurint will likely have similarly ill effects on migrants’ participating in civic life and may impede local governments’ ability to carry out their responsibilities. ⁴²

It is worth highlighting the relationship of this concern to the Reasonable Endorsement Test, which misses something important. Substate governments’ legitimacy rests in part on their responsibilities for community protection, health, safety, welfare, and public education. Members of substate polities can endorse their governments in part due to their abilities to fulfill such responsibilities. But where there is a workaround like Accurint that enlists substate governments in carrying out enforcement actions that undercut their ability to fulfill legitimating responsibilities, there is a kind of institutional incoherence at work. The result is that the grounds for endorsement of those substate governments are undercut. And that incoherence makes endorsement of a system like Accurint less reasonable. The relation between levels of government and reasonable endorsement is beyond the scope of this article, but certainly a question worth deeper exploration.

Conclusion

Here we want to step back and recognize the narrowness of our arguments so far. For one, the Lexis case can be explored and reframed in several ways. One might center the issues Accurint presents more thoroughly against surveillance frameworks canvassed in the critiques of data systems and immigration practices. However, the strength behind our approach is that it draws on explicitly applied ethics and provides a baseline condition for the justifiability of using certain technologies in a particular context. It does not purport to determine the only morally justified or morally best application of technology. Drawing on RCP's Reasonable Endorsement Test allows us to focus on one facet of the problem with using commercial databases in the vast immigration domain.

Our aim is to provide an avenue for scholars of border criminologies and migration ethics to integrate thinking about technology and justifiability into this complex set of issues. We also hope that the Lexis case and approach allow for further interdisciplinary engagement. We think it can mesh well with approaches from surveillance studies, STS, and immigration law. Future directions stemming from this project could include looking at moral justifications of technology use in connection with market incentives to surveil by private and state actors, how to address the ability of government actors to circumvent legal restrictions on information collection by using private data firms, and how certain technological uses may abdicate responsibility (as hinted by Molnar and our argument in the levels of government section).

Stepping back even further, there is a deeper issue about the relationship between the United States, individual states, and migrants. The entire immigration process—not just enforcement based on sophisticated technologies—deserves attention. It is plausible that the United States has failed to establish a fair and humane immigration system (e.g., by not providing the necessary channels for asylum seekers, by providing insufficient pathways for legal immigration, and by engaging in inhumane enforcement tactics). And this despite the fact that many people and industries in the United States rely on and are deeply connected to migrants.

Those broader issues are key to outlining fair terms of social cooperation. Yet we have not addressed them in this paper. And one might argue that these background issues make discussion of “reasonable endorsement” of Accurint's use in immigration enforcement too narrow to be useful. Perhaps. Nonetheless, the broader set of immigration issues is composed of myriad constituent issues, including use of sophisticated technologies in enforcement. If there are indeed some justifiable restrictions on migration, our sense is that how those restrictions are enforced (including with sophisticated technologies) is a worthwhile focus for evaluation.

Moreover, there are many different technologies that aid immigration enforcement that we cannot address here: biometric identification, facial recognition, motion sensors, location monitors, and more. Those raise overlapping (though discrete) sets of questions and are themselves important objects of moral analysis. Reasonable endorsement may be a useful way forward on those too.