US DHS’s Opaque Vetting Software Allegedly Relied on Poor-Quality Data and Discriminated against Immigrants

US Citizenship and Immigration Services (USCIS)’s ATLAS software used in vetting immigration requests was condemned by advocacy groups as a threat to naturalized citizens for its secretive algorithmic decision-making, reliance on poor quality data and unknown sources, and alleged discrimination of immigrants using biometric and sensitive information.

Risk classification

• Primary risk domain: 1 Discrimination & Toxicity
• Primary risk subdomain: 1.1 Unfair discrimination and misrepresentation

The ATLAS system explicitly permits the consideration of race, ethnicity, and national origin as screening criteria under certain circumstances, leading to the potential for unfair discrimination and disproportionate targeting of specific demographic groups.

Additional risk subdomains

• 7.3 Lack of capability or robustness: The system relies on shoddily maintained databases and historical fingerprint records, leading to automated red flags based on bureaucratic errors rather than actual fraud.

Causal factors

• Entity: AI
• Intent: Unintentional
• Timing: Post-deployment

The risk of wrongful denaturalization of innocent citizens arises as an unexpected and unintentional outcome of deploying the ATLAS system to pursue fraud detection.

EU AI Act risk tier

• Risk tier: 2 High Risk

Risk Level 2. High Risk: The ATLAS system is used by law enforcement and immigration authorities (DHS) to flag individuals for denaturalization and deportation, which has significant implications for fundamental rights and public interests, matching the High Risk category for law enforcement and migration control applications.

AI system and alleged parties

• AI system: ATLAS
• AI purpose: Threat Detection; Identification
• Behaviour type: Autonomous
• Alleged developer: US Citizenship and Immigration Services
• Alleged deployer: US Department of Homeland Security, US Citizenship and Immigration Services
• Alleged harmed parties: US naturalized citizens, US immigration applicants, US immigrants, US citizenship applicants

Harm severity

Severity ratings are based on this scale.

Highest direct severity in any category: Substantial. Severity is scored from Negligible to Catastrophic in each harm category, for harm the reports describe as caused directly or indirectly by the AI system.

• Physical: direct Negligible, indirect Negligible
• Infrastructure: direct Negligible, indirect Negligible
• Property: direct Negligible, indirect Negligible
• Financial: direct Negligible, indirect Negligible
• Environmental: direct Negligible, indirect Negligible
• Malicious content: direct Negligible, indirect Negligible
• Differential treatment: direct Substantial, indirect Negligible
• Civil rights: direct Substantial, indirect Substantial
• Democracy: direct Negligible, indirect Negligible
• Privacy: direct Substantial, indirect Negligible
• Psychological: direct Negligible, indirect Negligible
• Epistemic: direct Negligible, indirect Negligible
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Differential treatment

Reported: The report explicitly describes potential differential treatment based on race, ethnicity, and national origin.

Directly caused: The ATLAS system allows the consideration of race and ethnicity in 'exceptional instances' and country of origin as screening criteria, which advocates argue disproportionately targets individuals from Muslim-majority countries.

Inferred additional harm: The system's design likely leads to systemic bias and discriminatory enforcement patterns against naturalized citizens from specific demographic backgrounds.

Civil rights

Reported: The report explicitly describes threats to the civil rights of naturalized citizens, specifically the potential revocation of citizenship and deportation.

Directly caused: The ATLAS system automatically flags naturalized citizens for potential citizenship revocation and deportation, directly threatening their fundamental right to citizenship.

Indirectly caused: The system relies on unverified and potentially erroneous data from external databases without providing a mechanism for individuals to correct errors, violating basic due process.

Inferred additional harm: Thousands of naturalized citizens may face unlawful denaturalization and deportation due to automated bureaucratic errors, representing a severe violation of civil liberties.

Privacy

Reported: The report explicitly describes privacy assessments of the ATLAS system and its data-sharing capabilities.

Directly caused: The ATLAS system automatically scours and links data across multiple federal databases, including biometric data like fingerprints and social network linkages, to identify 'non-obvious relationships'.

Inferred additional harm: The continuous, automated surveillance and cross-referencing of millions of immigrants' personal and biometric data without their explicit consent or clear oversight constitutes a massive invasion of privacy.

People affected

• Occurrences reported: 1
• People reportedly harmed: 2628
• People reportedly exposed: 16500000

Potential causes

Management

• Prioritizing Deportation Over Accuracy: Agency leadership focused on maximizing referrals rather than system fairness.
• Inadequate Risk Assessments: Management failed to evaluate the risk of automated errors harming citizens.
• Expansion of Denaturalization Programs: Funding and offices were aggressively expanded to increase prosecutions.

Technology

• Secret Pattern-Based Algorithms: Secret predictive rules make determinations without external scrutiny.
• Linkage Analysis of Social Networks: System automatically maps relationships to infer potential criminal links.
• Autonomous Database Mining: Software autonomously scans federal databases to flag individuals for review.

Data Inputs

• Unverified Source Data: ATLAS presumes all database inputs are accurate, ignoring potential errors.
• Use of Biased Demographic Data: System considers race, ethnicity, and national origin in screening rules.
• Shoddy Fingerprint Records: Digitized historical fingerprint records may contain errors or mismatches.

Human Factors

• Over-Reliance on Automated Flags: Human reviewers triage system-generated notifications with minimal skepticism.
• Cognitive Bias in Investigations: ICE agents prioritize enforcement and tend to ignore conflicting evidence.
• Onus of Correction on Individuals: Victims must navigate complex federal agencies to correct algorithm source data.

Process and Methods

• Continuous Algorithmic Scrutiny: Post-adjudication screening subjects naturalized citizens to infinite checks.
• Rapid Four-Step Referral Process: Flowchart shows only four steps from automated flag to ICE referral.
• Lack of Algorithmic Transparency: DHS refuses to disclose the rules and criteria used by the rules engine.

Regulatory Environment

• Absence of Algorithmic Auditing: No independent regulatory oversight to evaluate ATLAS for disparate impact.
• Delayed Policy Reviews: Executive reviews of denaturalization policies missed public deadlines.
• Misuse of National Security Exemptions: Murky national security justifications shield the system from public disclosure.

Information quality

• Classification confidence: High
• Reason for confidence: The reports are highly detailed, drawing directly from government documents obtained via FOIA lawsuits. They provide specific metrics on the system's scale (16.5 million screenings, 120,000 flags) and clear descriptions of its operational mechanics, inputs, and consequences. While the exact algorithm remains secret, the systemic impacts and risks are well-documented.
• Ambiguities identified: The exact rules and predictive criteria used by the ATLAS algorithm are secret and undisclosed by the DHS.
• Alternative interpretations: The government may argue the system is merely an administrative tool to assist human decision-makers rather than an autonomous decision-making system.

The Department of Homeland Security's ATLAS system represents a significant automation of immigration enforcement, conducting millions of autonomous screenings that flag naturalized citizens for potential denaturalization. While intended to detect fraud, the system's reliance on flawed databases and lack of transparency poses substantial risks to societal stability, civil liberties, and the integrity of sovereign administrative processes.

• Overall national security impact: Substantial
• Scope: Single nation
• Primary target: United States
• Alleged perpetrator: United States Government

Threat characteristics

• Imminence: Long-term. Represents an ongoing, long-term systemic policy and administrative concern rather than an immediate, acute crisis.
• Autonomy: Human-supervised. The system autonomously scans databases and generates red flags, but final legal actions require human-initiated paperwork and court processes.
• Novelty: Evolved capability. Represents a significant scaling and automation of existing database cross-referencing and immigration screening practices.

Impact by dimension

• Physical security: Negligible. No physical security threats, kinetic attacks, or critical infrastructure compromise reported in connection with the system.
• Information security: Minor. Utilizes and cross-references federal intelligence databases, but does not involve intelligence compromise, classified data theft, or active foreign information warfare.
• Sovereignty: Substantial. Systemic integration of an opaque algorithmic tool into core immigration and border control functions, affecting government decision-making processes regarding citizenship.
• Economic security: Negligible. No significant threats to national economic stability, strategic industries, or technological competitive advantage identified.
• Societal stability: Substantial. Enables mass algorithmic surveillance of millions of naturalized citizens, risking systemic discrimination and severe civil rights violations through erroneous denaturalization filings.