IRS Audited Black Taxpayers More Frequently Reportedly Due to Algorithm

The IRS was auditing Black taxpayers more frequently than other groups allegedly due to the design of their algorithms, focusing on easier-to-conduct audits which inadvertently correlated with the group's pattern of tax filing errors.

Risk classification

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

The IRS audit selection algorithm results in unfair discrimination and unequal outcomes, disproportionately targeting Black taxpayers for audits at rates 2.9 to 4.7 times higher than non-Black taxpayers.

Additional risk subdomains

• 1.3 Unequal performance across groups: The algorithm's effectiveness in identifying actual tax underreporting magnitude differs by demographic group, performing worse at finding high-value evasion when targeting lower-income Black taxpayers.

Causal factors

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

The racial disparity was an unintended consequence of the IRS's deployed audit selection algorithm, which was designed to optimize cost-efficiency and detect underreporting.

EU AI Act risk tier

• Risk tier: 2 High Risk

High Risk: The system is used by a government agency for resource allocation and enforcement (tax audits), which directly impacts the fundamental rights and financial well-being of citizens, fitting the category of high-risk AI systems used in law enforcement or essential public services.

AI system and alleged parties

• AI system: IRS audit selection algorithm
• AI purpose: Resource Allocation; Financial Processing
• Behaviour type: Autonomous
• Alleged developer: Internal Revenue Service
• Alleged deployer: Internal Revenue Service
• Alleged harmed parties: Black taxpayers

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 Substantial
• Civil rights: direct Substantial, indirect Negligible
• Democracy: direct Negligible, indirect Negligible
• Privacy: direct Negligible, 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 systemic differential treatment where Black taxpayers are audited at 2.9 to 4.7 times the rate of non-Black taxpayers.

Directly caused: The IRS audit selection algorithm directly caused Black taxpayers to be audited at significantly higher rates, particularly single Black men with dependents claiming the EITC, who were audited at 7.73% compared to 3.46% for non-Black claimants.

Indirectly caused: The algorithm's focus on simple, low-cost correspondence audits indirectly led to a disproportionate burden on lower-income Black taxpayers who are less likely to report business income.

Inferred additional harm: This differential treatment likely exacerbated systemic racial wealth disparities by disproportionately clawing back tax credits from Black families.

Civil rights

Reported: The report explicitly describes racial disparities that raise civil rights concerns, noting that the findings confirm stark racial disparities exist.

Directly caused: The algorithm's biased selection process violated the principle of equal treatment under the law, disproportionately burdening Black taxpayers based on race-blind but proxy-heavy criteria.

Inferred additional harm: The systemic targeting of a protected racial group for enforcement actions represents a broader civil rights issue, potentially violating federal protections against discriminatory impacts in government programs.

People affected

• Occurrences reported: 1
• People reportedly harmed: 167700
• People reportedly exposed: 148000000

Potential causes

Management

• Secret Selection Algorithms: Keep selection algorithms confidential, preventing external equity reviews.
• Resource Allocation Choices: Management prioritized cheap, simple audits over high-value complex ones.
• Lack of Equity Assessment: Management did not proactively analyze audit algorithms for disparate impacts.

Technology

• Algorithmic Selection Bias: Algorithm prioritizes audit likelihood over the magnitude of underreporting.
• Targeting Refundable Credits: Algorithm targets refundable credit errors rather than total tax evasion.
• Dependent Database Flags: Database automatically flags potential issues and generates mail audits.

Data Inputs

• Lack of Race Data: IRS does not collect race data, preventing detection of systemic bias.
• Race-Blind Model Assumption: Assuming race-blind inputs prevent racially biased algorithmic outcomes.
• Simplified Return Focus: Algorithm focuses on simple returns with easy-to-identify errors.

Human Factors

• Assumption of Neutrality: Staff assumed race-blind algorithms would produce equitable outcomes.
• Overreliance on Automation: Officials relied on automated systems due to resource constraints.

Process and Methods

• Cost-Saving Audit Selection: IRS avoids auditing business income returns because they cost more to audit.
• Focus on Correspondence Audits: Mail audits are prioritized over complex, resource-intensive field audits.
• Minimizing No-Change Rate: Process designed to minimize audits that result in no tax change.

Regulatory Environment

• Budget Cuts and Constraints: Congressional budget cuts forced IRS to rely on cheap automated audits.
• Lack of Algorithmic Oversight: No regulatory requirements existed to audit federal algorithms for racial bias.
• Congressional Focus on Credits: Congressional pressure to prevent over-claiming of refundable tax credits.

Information quality

• Classification confidence: High
• Reason for confidence: The reports are based on a rigorous, peer-reviewed scientific study conducted by major academic institutions (Stanford, Michigan, Chicago) and the Treasury Department, analyzing a massive dataset of 148 million tax returns. The findings are highly consistent across all sources.
• Ambiguities identified: The exact proprietary algorithms used by the IRS remain secret ('likened to the recipe for Coca-Cola'), requiring researchers to model and infer the exact mechanisms driving the disparity.
• Alternative interpretations: None. The statistical disparity is robustly demonstrated, and the IRS itself acknowledged the need to address the findings.

An IRS audit selection algorithm was found to disproportionately target Black taxpayers due to technocratic design choices that optimized for low-cost audits. While raising significant civil rights and administrative equity concerns, the incident has minor national security implications and is manageable through standard domestic policy and algorithmic reforms.

• Overall national security impact: Minor
• Scope: Single nation
• Primary target: United States
• Alleged perpetrator: Internal Revenue Service

Threat characteristics

• Imminence: Long-term. The issue represents an ongoing, systemic policy and administrative concern rather than an immediate national security crisis.
• Autonomy: Human-supervised. The algorithm autonomously flags and selects tax returns for audits, but human IRS administrators oversee the system and execute the audits.
• Novelty: Established threat. Systemic algorithmic bias and disparate impact in public sector resource allocation are well-documented and established issues.

Impact by dimension

• Physical security: Negligible. No threat to physical systems, critical infrastructure, or human safety is indicated in the incident report.
• Information security: Negligible. No compromise of intelligence capabilities, classified models, or systematic information warfare operations occurred.
• Sovereignty: Minor. Core government functions (tax administration) were affected by systemic algorithmic bias, but the issue is manageable within standard administrative and policy procedures.
• Economic security: Negligible. While individual taxpayers faced financial burdens, the incident did not threaten macro-economic stability, strategic industries, or technological competitive advantage.
• Societal stability: Minor. The algorithm resulted in systemic racial disparities, disproportionately auditing Black taxpayers. This raises civil rights concerns but does not threaten overall national societal stability.