Report: U.S. Government Is Using an Algorithm to Flag American Citizens for Denaturalisation

Report: U.S. Government Is Using an Algorithm to Flag American Citizens for Denaturalisation
U.S. Immigration and Customs Enforcement (ICE) officers prepare for immigration raids in April 2018 in New York City. (Photo: John Moore, Getty Images)

U.S. citizens can be kicked out of the country based on the findings of a secret algorithm. The U.S. Department of Homeland Security is using an Amazon-hosted system called ATLAS that analyses millions of records and can be used to automatically flag naturalised Americans for the revocation of their citizenship, the Intercept reported this week.

ATLAS is part of the U.S. Customs and Immigration Service’s (UCIS) Fraud Detection and National Security Data System (FDNS-DS) and runs on Amazon Web Services servers. Its official purpose is to compare case records from the immigration system to other federal databases, looking for indications of criminal, dishonest, or dangerous behaviour, as well as inconsistencies that authorities might see as evidence of fraud or using multiple identities.

Documents obtained by the Open Society Justice Initiative and Muslim Advocates through Freedom of Information Act (FOIA) requests and shared with the Intercept show that the system is ultimately designed with the endgame of deportation in mind, and it provides tools that could help immigration authorities go after naturalised citizens based on decades-old administrative mistakes.

The Intercept wrote that denaturalisation was once very rare, but under George W. Bush’s administration, the DHS digitised fingerprint data and identified some 1,029 people who it accused of evading deportation orders and going on to become naturalised citizens anyway. Barack Obama’s administration, no stranger to devastating mass deportations, did subsequently urge officials only to strip the citizenship of those posing a clear danger, but his openly racist successor Donald Trump ramped the machinery of the denaturalisation machine back up. The Justice Department announced it intended to refer some 1,600 naturalised citizens for prosecution under his tenure in 2018, according to The Intercept.

The DOJ requested $US207.6 ($286) million in the 2019 and 2020 budgets to pursue hundreds of additional leads, review 700,000 immigration records under a similar program, and create an office devoted to stripping citizenship from those accused of lying their way through the process. USCIS documents obtained by the Open Society Initiative and shared with the Intercept showed that as of April 2020, USCIS had filed paperwork related to denaturalisation in at least 2,628 cases, of which 745 were pending and 502 had been referred to the DOJ.

According to The Intercept, documents show ATLAS analyses information including biometrics like fingerprints, as well as draws information from databases including the FBI’s terrorism watchlist and the National Crime Information Centre, which have often “been criticised as being poorly managed.” In what a 2020 privacy document describes as “exceptional instances,” the system may also take race and ethnicity into account when making determinations.

Another 2016 privacy assessment of FDNS-DS showed ATLAS can also flag individuals based on their known associates, stating it has the capability to identify “linkages or relationships among individuals to assist in identifying non-obvious relationships… with a potential nexus to criminal or terrorist activities.” Some of the information is classified.

It’s hard not to see the use of a secret algorithm as merely a way to shield the process from outside scrutiny, such as whether infamously unreliable sources of information like gang databases or data-harvesting contractors factor into ATLAS decisions. The Intercept wrote that DHS refuses to clarify exactly how the system works or what data points it uses to flag immigrants for potential revocation of citizenship, but the documents indicate that USCIS can cross-check immigrants’ records against ATLAS in a wide variety of situations:

Immigrants come into contact with ATLAS, according to the 2020 privacy assessment, when one “presents him or herself” to the USCIS for some reason, of which there are many; when “new derogatory information is associated with the individual in one or more U.S. Government systems”; or, according to the 2016 privacy document, whenever “FDNS performs an administrative investigation.” This apparently can happen even after an immigration-related decision has been made: Among the FOIA documents shared with The Intercept is a USCIS memo noting that ATLAS is used to detect “fraud patterns in immigration benefit filings … either pre- or post-adjudication,” suggesting that an immigrant could be subjected to algorithmic scrutiny indefinitely after their filing is approved.

The 2020 document describes that ATLAS “contains a rules engine that applies pattern-based algorithms to look for indicators of fraud, public safety, and national security concerns,” which is described as “predictive.” DHS doesn’t disclose the way ATLAS’s algorithm works or what data points are ultimately used to generate red flags, though the 2020 document claims that there are rules “limiting the consideration” of individual’s connection solely by birth or citizenship to another country unless there is a need “based on an assessment of intelligence and risk and in which alternatives do not meet security needs.”

According to the Intercept, a USCIS spreadsheet from 2020 obtained via FOIA shows 12 categories of ATLAS alert, including notifications that appear to refer to individuals protected by the Deferred Action for Childhood Arrivals (DACA) program, the Department of Defence, national security categories, and “multiple identities.” The 2020 privacy documents note that USCIS presumes all data submitted to it is accurate and places the onus of the responsibility for correcting bad data on the individual being screened.

The documents instead say such corrections require the individual under investigation to reach out to an administrator of the original database, the Intercept wrote. For example, an individual seeking to change the outcome of an ATLAS review that fast-tracked them for denaturalisation based on an erroneous entry in the FBI’s terrorism or criminal databases would first have to determine the nature of the error and then appeal directly to the FBI to fix it.

The same would apply to someone who had inconsistencies on old paperwork thanks to shoddy work by attorneys or translators, or who did everything correctly but had immigration authorities screw up handling or analysis of decades-old fingerprint records. Given that the staggering lack of transparency within the U.S. immigration system, the vulnerability of those adversely targeted by it, and the possibility the original source information might be classified, the likelihood of contesting such mistakes before they snowball into full-blown removal proceedings seems slim.

When ATLAS makes a negative determination, the Intercept wrote, it sends out a “System Generated Notification” that is “triaged” and sent to FDNS-DS if potentially “actionable.” A human review at FDNS-DS is then used to identify whether it can carry out “possible criminal denaturalisation referral” and refer the flagged individual as a possible criminal suspect to U.S. Immigration and Customs Enforcement, which is well known for prioritising aggressive enforcement above actually bothering to ensure it’s working with accurate information.

According to the LA Times, ICE regularly sucks citizens who have done nothing wrong into its dragnet based on shoddy investigations and has faced dozens of lawsuits for false imprisonment based on “cursory computer searches,” with agents often never bothering to gather or deliberately ignoring evidence to the contrary like passports, interviews, and conflicting records.

In 2019 alone, according to The Intercept, ATLAS ran 16.9 million screenings and generated 120,000 red flags on the basis of suspected fraud or “threats to national security and public safety.” One USCIS flowchart shows as few as four steps can be involved between an ATLAS flag and referral to ICE.

Individuals who are denaturalised are not always removed from the country. But The Intercept found that numerous USCIS and ICE documents indicated the agencies’ preferred outcome is deportation, with one 2009 ICE memo instructing agents to pursue detention and removal in cases of suspected identity or benefits fraud even when the DOJ has declined to criminally prosecute. One USCIS spreadsheet showed that in 2018 and 2019, authorities rejected 10 out of 10 settlement proposals that included protections against deportation.

“The whole point of ATLAS is to screen and investigate so that the government can deny applications or refer for criminal or civil or immigration enforcement,” Deborah Choi of Muslim Advocates told The Intercept. “The purpose of the secret rules and predictive analytics and algorithms are to find things to investigate.”

The Open Society Justice Initiative has filed another FOIA with DHS and USCIS seeking to force them to disclose the algorithm that powers ATLAS. Joe Biden’s administration began a review of the denaturalisation program in February 2021, but hasn’t so much as publicly acknowledged it since blowing past a May deadline without publishing any findings.

Amazon, which has faced employee and activist uproar over its deals with immigration authorities in recent years, was listed as ATLAS’s web host as of 2020, according to the report. The company didn’t return the Intercept’s request for comment.