Major Universities Are Using Race as a “High Impact Predictor” of Student Success

Several major universities are using a tool that uses race as one factor to predict student success.

Risk classification

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

The EAB Navigate system explicitly uses race as a predictor, resulting in Black and Latinx students being disproportionately labeled as high risk and unfairly steered away from competitive majors.

Additional risk subdomains

• 5.1 Overreliance and unsafe use: Overworked academic advisors rely on the proprietary risk scores to make critical educational recommendations without understanding how the scores are calculated.
• 7.4 Lack of transparency or interpretability: The risk scoring algorithm is proprietary and protected, leaving university administrators and advisors unable to explain or verify the decision-making process.

Causal factors

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

The discriminatory risk scores and subsequent steering of minority students are caused by the outputs of the deployed EAB Navigate AI system, which was an unintended consequence of trying to improve student retention.

EU AI Act risk tier

• Risk tier: 2 High Risk

High Risk: The system is used in education ('Educational and vocational training systems affecting access to education') to evaluate students and predict success, which directly affects their academic trajectory and access to opportunities.

AI system and alleged parties

• AI system: Navigate advising software (EAB)
• AI purpose: Behavioral Modeling; Resource Allocation
• Behaviour type: Assistant
• Alleged developer: EAB
• Alleged deployer: University of Wisconsin–Milwaukee, University of Massachusetts Amherst, University of Houston, Texas A&M University, more than 500 colleges, Georgia State University
• Alleged harmed parties: Latinx college students, indigenous students, Black college students

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 Minor, indirect Minor
• Environmental: direct Negligible, indirect Negligible
• Malicious content: direct Negligible, indirect Negligible
• Differential treatment: direct Substantial, indirect Substantial
• 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

Financial

Reported: Yes, the report mentions Georgia State University spent $318,000 on its EAB subscription for 2020.

Directly caused: Georgia State University paid $318,000 for its EAB subscription in 2020.

Indirectly caused: Students steered away from STEM majors face potential long-term reductions in lifetime earnings, though this is not quantified in the report.

Inferred additional harm: Across the 500 universities using Navigate, the collective financial cost of subscriptions likely exceeds $100,000,000.

Differential treatment

Reported: Yes, the report explicitly describes Black and Latinx students being disproportionately labeled as high risk compared to their White peers.

Directly caused: Black students were deemed high risk at up to quadruple the rate of their White peers across multiple universities.

Indirectly caused: Advisors steered underrepresented students in STEM to other 'lower risk' majors based on these racially biased scores.

Inferred additional harm: Widespread systemic steering of minority students across 500 universities, reinforcing historical educational disparities.

Civil rights

Reported: Yes, the practice of using race as a predictor in educational evaluations raises civil rights concerns regarding equal protection and discrimination.

Directly caused: The algorithmic system uses race as a 'high-impact predictor' to assign risk scores, violating principles of fair and equal treatment in public education.

Indirectly caused: Steering minority students away from competitive majors based on racially biased scores limits their educational opportunities.

Inferred additional harm: Potential systemic violations of civil rights laws (such as Title VI of the Civil Rights Act) across hundreds of public universities using this software.

Privacy

Reported: Yes, students were scored by algorithms using their personal data (race, GPA, test scores) without their knowledge or consent.

Directly caused: Universities fed student demographic and academic data into proprietary third-party algorithms without notifying the students.

Inferred additional harm: Millions of students' private educational and demographic records have been processed by EAB's proprietary models without explicit student consent.

People affected

• Occurrences reported: 1
• People reportedly harmed: 5000
• People reportedly exposed: 120000

Potential causes

Management

• Prioritizing Retention over Equity: Schools adopt software to preserve tuition revenue amidst funding cuts.
• Lack of Algorithmic Due Diligence: Administrators buy software without understanding its underlying variables.
• Underinvestment in Student Support: Buying software instead of funding actual tutors, advisers, or micro-loans.

Technology

• Proprietary Black-Box Algorithms: Proprietary nature prevents universities from auditing bias in risk scores.
• Historical Bias Encoding: Models trained on old data encode past discriminatory practices.

Data Inputs

• Race as a Risk Predictor: Using race directly as a predictor leads to racially disparate risk scores.
• Omission of Financial Data: Failing to include financial struggles, a primary reason for dropping out.
• Biased Historical Student Data: Using 2 to 10 years of historic outcomes data encodes systemic disparities.

Human Factors

• Adviser Bias and Steering: Advisers steer minority students to 'easier' majors based on risk scores.
• Overreliance on Quick Snapshots: Overworked advisers rely on 30-second dashboards instead of deep review.
• Lack of Student Awareness: Students are judged by algorithms without their knowledge or consent.

Process and Methods

• Inadequate Adviser Training: Universities provide little guidance on how to interpret or use risk scores.
• High Student-to-Adviser Ratios: Ratios up to 700-to-1 force reliance on automated risk scores.
• Inappropriate Use of Risk Explorer: Software features encourage pushing students out of preferred majors.

Regulatory Environment

• Lack of Algorithmic Auditing: No regulations require independent bias testing for education software.
• Proprietary IP Protections: Trade secret protections shield the algorithms from public scrutiny.

Information quality

• Classification confidence: High
• Reason for confidence: The report provides detailed, data-backed evidence from public records requests and interviews with university administrators, students, and experts. The role of the AI system, the specific variables used (race), and the resulting disparities are clearly documented across multiple institutions.
• Ambiguities identified: The exact proprietary formulas of EAB's algorithms remain undisclosed, and some universities redacted their model summaries.
• Alternative interpretations: EAB claims the tool is intended to highlight existing disparities to prompt institutional support rather than to steer students, though the practical effect reported is steering.

The incident involves the widespread use of EAB's Navigate software in US universities, which uses race as a predictor to assign risk scores, disproportionately labeling minority students as 'high risk'. While raising significant concerns regarding algorithmic bias, civil rights, and educational equity, the national security implications are minor and manageable through standard civil rights enforcement and educational policy.

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

Threat characteristics

• Imminence: Long-term. Represents an ongoing, systemic issue embedded in post-deployment university software rather than an active, rapid-onset national security crisis.
• Autonomy: Human-supervised. The AI acts as an advising assistant, generating risk scores that human advisors interpret and act upon to steer students.
• Novelty: Evolved capability. Algorithmic bias and training on historical data are established issues, but applying proprietary predictive scoring to steer university students represents an evolved scale of demographic profiling.

Impact by dimension

• Physical security: Negligible. No physical systems, critical infrastructure, kinetic attacks, or threats to human safety are involved in this academic advising software incident.
• Information security: Negligible. No indications of intelligence compromise, classified data theft, foreign disinformation campaigns, or state-sponsored media manipulation.
• Sovereignty: Negligible. This is a domestic educational administration issue. It does not threaten state authority, territorial control, electoral systems, or core sovereign government operations.
• Economic security: Minor. May influence long-term workforce demographics in STEM fields, but does not represent a direct threat to national financial stability, critical supply chains, or strategic technology.
• Societal stability: Minor. Systemic racial bias in educational software affects student opportunities and civil rights, though manageable through standard legal, regulatory, and educational policy frameworks rather than national security intervention.