Under the Biden administration, Immigration and Customs Enforcement hasn’t stopped arresting and detaining people, tearing apart immigrant communities, families, and livelihoods across the country. Advocates are pushing the administration to reverse course on some of the agency’s grossest abuses, many of which were brought to mainstream attention throughout the Trump era. But the reality is still dark: ICE’s national, law enforcement-supported network of detention facilities, where more than 22,000 immigrants are currently incarcerated, and its ballooning electronic monitoring program, where close to 132,000 people are constantly surveilled, remain stronger than ever.
As part of my work with Just Futures Law and Mijente, we documented the latter program at length in a report that called for its immediate end. And that’s because, as in the criminal legal context, ICE’s so-called Alternatives to Detention Program is no more than a technological extension of incarceration — one that harms people in its clutch and expands systems of carceral control in destructive ways. Our work in shedding light on and dismantling ICE’s digital prisons continues. And crucial to that broader goal is a greater public understanding of other discriminatory tools the agency uses to cage and criminalize immigrants. Among these technologies is ICE’s Risk Classification Assessment, an algorithmic tool that remains an integral part of perpetuating policies that differentiate between “safe” and “dangerous” immigrants — and feeds ICE’s electronic monitoring program at the center of our work.
Again, as in the criminal context, those distinctions, which politicians and even some liberal advocates have latched onto in policy debates over “threats” to public safety, are harmful and fall hardest on Black and Brown immigrants. So understanding how ICE has used this false distinction in its detention and other policy decisions remains a cornerstone of immigrants’ rights advocacy. A greater grasp of how ICE relies on notions of predicting future “risk” in deciding whether to release or cage a person matters because these tactics are one piece of ICE’s enforcement and deportation machine. Challenging this tool now is especially urgent, as Congress has expressed renewed interest in expanding ICE’s risk assessment analysis system in its appropriations process.
And as with the demands to “free them all” that grew louder as COVID-19 hit detention facilities, nothing short of ending ICE’s use of algorithmic tools will get us closer to the goal of ending all forms of immigration imprisonment.
The Life and Times of a Flawed Algorithm
ICE developed the RCA between 2009 to 2012 in response to then-ICE Director Dora Schriro’s recommendation to “institute a nationwide risk assessment tool.” The agency contracted IBM for $7.6 million to create the Automated Threat Prioritization, which formed the technical foundation of the tool. When the RCA was first being implemented, many policymakers and some advocates supported the measure as a form of progress, arguing that it would limit ICE’s detention capabilities “based on an assessment of who must be incarcerated.” Since 2013, ICE has used the RCA nationally, inputting various metrics from hundreds of thousands of newly arrested individuals yearly in its algorithmic assessment to determine the “risk” they pose. Information entered in this algorithm includes biographical information about the person, where and how they were arrested by ICE, and any “special vulnerabilities,” such as physical or mental health concerns the detainee may have. On the ground and in practice, ICE relies most heavily on a supposed “risk to public safety” and “risk of flight” in its detention determinations.
Yet early on, the Department of Homeland Security’s own internal watchdog identified problems, noting in no uncertain terms that this risk assessment tool was “time consuming, resource intensive, and not effective in determining [whom] to release or under what conditions.” A first-of-its-kind national study on the RCA’s impact strongly suggested that the tool, which overwhelmingly recommended detention over release, did little more than add a “false veneer of scientific analysis” to a system predisposed to over-detention in the first place. The first year of its national implementation by the Obama administration confirmed that of all the individuals run through the RCA, the tool recommended for release a mere 1 percent of those subject to it, while 91 percent ended up detained. And further sunlight into the algorithm demonstrated how the RCA was rigged from the start.
While one significant obstacle to releasing people from ICE detention is federal immigration law’s requirement that certain immigrants be placed in mandatory detention, ICE has fought for the broadest legal interpretation of this category across administrations. Moreover, DHS itself has reported that 63% of people who are not subject to mandatory detention and had a special vulnerability that counseled their release were still ordered detained. Between 2017 to 2019, in the jurisdiction of ICE’s New York field office alone, more than 97% of people were detained when the Trump administration, as documents disclosed in ongoing litigation revealed, all but eliminated the ability for the tool to recommend release; by contrast, from 2013 to June 2017, in that same jurisdiction, roughly 47% of people the government deemed low-risk were released.
Garbage In, Garbage Out
That the RCA is subject to political manipulation in line with detention goals is no surprise to those who have closely examined it. Thanks to invaluable Freedom of Information Act litigation and requests, scholars from the University of Maryland and Duke University uncovered a trove of ICE records that allowed them to piece together how the RCA operates, and much of what we know today about the tool would simply not be possible without their work. The data points fed into the algorithm dictate much of what the machine puts out — and how much of it is manipulated to recommend detention, as detention is what ICE does best.
The data points fed into the algorithm dictate much of what the machine puts out — and how much of it is manipulated to recommend detention, as detention is what ICE does best.
ICE analyzes “risk to public safety” by ranking previous charges or convictions by their “severity,” which then get assigned a different score as outlined by the RCA rubric. As thoroughly documented in the FOIA study by Professors Kate Evans and Robert Koulish, ICE has full discretion to alter the severity score for a broad range of charges, depending on enforcement priorities at any given time. In a criminal legal system already stacked against Black and Brown people, the charges or convictions an individual has on their record, no matter if ultimately adjudicated, can be easily skewed. This is especially true for charges that are pending or dropped, which ICE considers as risk factors in the RCA. Even if we pretended for a moment that charges or convictions associated with an individual were actually helpful in predicting future criminal activity, which a number of studies have shown is not the case, ICE remains a civil enforcement agency, primarily focused on civil immigration detention.
Moreover, the score for “risk to public safety” factors in ICE’s reliance on other databases plus an RCA interview conducted by an ICE officer. These databases may include the National Crime Information Center database, biometric information-sharing databases, and so-called gang databases. Each of these databases imports information from various sources that are riddled with inputs obtained from racist practices in how the data is collected. For example, gang databases are overwhelmingly composed of Black and Latinx youth who are targeted for the colors they wear, whom they associate with, their tattoos, and where they live. Inputting a flag from a gang database into the RCA allows the algorithm to recreate and proliferate the same targeting of Black and Brown communities that landed them in the database to begin with — and compound the very real carceral harm of this type of guilt by association.
As Evans and Koulish put it in their research, ICE’s policy preferences for carceral inputs meant that “the RCA’s algorithm lost the ability to measure true risk,” turning instead into “a tool of prosecution, pairing detention with enforcement preference regardless of risk.”
An individual’s supposed “risk of flight,” meanwhile, is next in the RCA assessment, and it raises its own concerns. This metric, which purports to assess whether someone will show up for court, is based on the ICE officer’s interview with the person. Data points for this part of the assessment include whom the individual lives with, how long they have lived there, whether they have a U.S. citizen spouse or child, and their level of family or community support, among other inputs. Generally, though, a low flight-risk score remains elusive for many migrants, in no small part because ICE retains full discretion to game many of the factors that would ensure a lower score. (Even if we accepted that “flight risk” could be a real problem among those caught in the government’s deportation machine, the weight of available evidence shows that the vast majority of immigrants and asylum seekers who aren’t detained do show up for their court dates.)
Worse yet, the RCA has been anything but consistent, with changing values for each data point over time. These changes have often been tied to prosecutorial discretion memos that determine ICE’s priority targets for immigration enforcement, which changes with every administration. Consistently since 2014, when the Obama administration announced it would be targeting those who were charged for, among other things, drug distribution and significant misdemeanors, the algorithm has gotten more stringent, penalizing people more harshly for previous charges and convictions. Under the Trump administration, things predictably got more punitive for migrants, and the RCA was changed again so that even those assigned low-risk scores — like people without criminal records — were detained without bond anyway.
Risk Assessments Perpetuate Incarceration and Profit-Making
Why did an algorithm that was supposedly designed to limit detention as a final resort continue to detain people in such large percentages? Because as data from across sectors shows, risk assessment tools are built to justify incarceration, not to reduce it, let alone eliminate our reliance on it. As implemented, the RCA reinforces the narrative that immigrants need to be assessed for their dangerousness, and their mere presence in this country criminalized. It lends an air of legitimacy to ICE’s historic mass detention and deportation strategy, all the while pretending that its detention decisions are impartial and based in science. As Evans and Koulish concluded, the RCA’s failure across administrations “on such a dramatic scale, and the resulting harm to hundreds of thousands of people, calls into question the legitimacy of detention of any immigrant.”
ICE's risk assessment tool lends an air of legitimacy to the agency's historic mass detention and deportation strategy, all the while pretending that its detention decisions are impartial and based in science.
Of course, underpinning this house of cards is the reality that “risk” is good for the incarceration industry. The higher percentages of people that are recommended by ICE’s algorithm for detention, the more beds are filled in detention centers. With mandatory bed numbers established by Congress, a significant result of lobbying by corporate behemoths like the GEO Group, ICE has no reason to use the RCA to release people when its ultimate goal is mass detention. The tool is yet another way for ICE and companies like GEO Group to profit off the backs of Black and Brown immigrants. The same is true for the growing universe, under the Biden administration, of people who are being placed on ICE’s electronic monitoring program, a dangerous technological extension of physical detention that perpetuates the same, profit-making status quo.
The RCA, like just about all risk assessment tools, serves the premise that Black and Brown people are threats to the state and amorphous notions of public order. From algorithms that are used by children’s services agencies that break apart families to predictive policing that pushes police back into neighborhoods that they are already targeting, risk assessments are another way of continuing racist policing and incarceration practices hiding behind the curtain of “data-driven” technology. But the truth is that “risk” has always been a racialized metric to punish Black, Brown, and poor communities, and ICE’s algorithm does just that. It won’t be until the agency frees everyone from the reach of its oppressive tactics, systems, and facilities that immigrant communities will truly be free.