VA Suicide Prevention Algorithm REACH VET Reportedly Prioritizes Men Over Women Veterans

An AI program named REACH VET, designed and used by the Department of Veterans Affairs (VA) to prevent veteran suicides, was reportedly found to prioritize white men while neglecting female veterans and survivors of military sexual trauma. This oversight persists despite rising suicide rates among these groups. The incident is an example of algorithmic bias and the exclusion of critical risk factors for female veterans.

Risk classification

• Primary risk domain: 1 Discrimination & Toxicity
• Primary risk subdomain: 1.3 Unequal performance across groups

The REACH VET algorithm performs unequally across demographic groups, failing to accurately identify suicide risk in female veterans because it excludes risk factors specific to them.

Additional risk subdomains

• 7.3 Lack of capability or robustness: The model lacks the capability to reliably predict suicide risk for the female veteran population due to missing critical clinical variables.

Causal factors

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

The systemic bias and exclusion of female risk factors was an unintentional outcome of the REACH VET algorithm's design and deployment.

EU AI Act risk tier

• Risk tier: 2 High Risk

High Risk: The REACH VET system is an AI application used in healthcare to identify clinical suicide risk, which has significant implications for safety and fundamental rights.

AI system and alleged parties

• AI system: REACH VET
• AI purpose: Health AI Assistant; Resource Allocation
• Behaviour type: Assistant
• Alleged developer: Department of Veterans Affairs (VA)
• Alleged deployer: Department of Veterans Affairs (VA)
• Alleged harmed parties: Veterans, Survivors of military sexual trauma, Female veterans

Harm severity

Severity ratings are based on this scale.

Highest direct severity in any category: Minor. 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 Minor, indirect Substantial
• Civil rights: direct Minor, indirect Negligible
• Democracy: direct Negligible, indirect Negligible
• Privacy: direct Negligible, indirect Negligible
• Psychological: direct Minor, indirect Negligible
• Epistemic: direct Negligible, indirect Negligible
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Differential treatment

Reported: Yes, the report explicitly describes systemic differential treatment where the algorithm prioritizes male veterans over female veterans.

Directly caused: The REACH VET algorithm considers variables like 'divorced and male' and 'widowed and male' but excludes female-specific risk factors like military sexual trauma and intimate partner violence.

Indirectly caused: Female veterans are deprioritized for proactive suicide prevention outreach, receiving less support relative to their actual risk levels.

Inferred additional harm: Systemic unequal access to life-saving mental health resources for the growing population of female veterans.

Civil rights

Reported: Yes, the report mentions potential violations of civil rights and executive orders prohibiting AI discrimination.

Directly caused: The algorithm's bias potentially violates executive orders requiring federal agencies to ensure AI models do not perpetuate discrimination.

Inferred additional harm: Systemic discrimination against female veterans in federal healthcare delivery, violating equal protection principles.

Psychological

Reported: The report describes distress and feelings of being marginalized among female veterans upon learning about the algorithm's bias.

Directly caused: Female veterans like [PERSON_002] expressed feeling pushed to the back again and experienced emotional distress, tearing up when informed of the bias.

Inferred additional harm: Widespread feelings of neglect, anxiety, and lack of trust in VA healthcare services among the 800,000 female veterans.

People affected

• Occurrences reported: 1
• People reportedly exposed: 800000

Potential causes

Management

• Failure to Address AI Bias: Management failed to establish compliance reviews and monitors for AI.
• Rationalizing Gender Exclusion: Leadership defended the exclusion of military sexual trauma from the model.

Technology

• Biased Algorithm Design: The REACH VET model favors majority demographics like white men.
• Exclusion of Female Risk Factors: Algorithm excludes factors like pregnancy, menopause, and firearm ownership.

Data Inputs

• Underreported Trauma Data: Military sexual trauma is heavily underreported in formal VA screenings.
• Omission of Critical Variables: Algorithm's 61 variables omit intimate partner violence and sexual trauma.

Human Factors

• Systemic Bias Against Women: Historically, the VA has habitually overlooked female veteran needs.
• Overreliance on Statistical Majority: Developers assumed predictive power for men applies equally to women.

Process and Methods

• Flawed Feature Selection: Officials excluded sexual assault history as not statistically powerful.
• Lack of Tailored Risk Models: VA failed to build separate, tailored suicide risk models for women.

Regulatory Environment

• Weak Compliance Enforcement: VA did not actively implement presidential executive orders on AI bias.
• Lack of Public Transparency: VA declined to share outreach dashboard data, limiting external oversight.

Information quality

• Classification confidence: High
• Reason for confidence: The reports provide detailed accounts from VA officials, independent technology experts, and veterans, with specific statistics on suicide rates and the variables used by the algorithm.
• Ambiguities identified: The exact statistical weight of the 61 variables is not fully detailed, and the direct causal link between the algorithm's bias and the 24% spike in female suicides is not statistically proven in the text, only suggested as a potential factor.
• Alternative interpretations: The exclusion of military sexual trauma could be interpreted as a purely statistical decision based on predictive power rather than systemic bias, as argued by some VA officials.

The Department of Veterans Affairs' REACH VET suicide prevention algorithm exhibits systemic bias by prioritizing male veterans and failing to account for key female-specific risk factors. While this raises serious ethical, civil rights, and clinical concerns regarding federal healthcare delivery, its direct national security impact is minor, representing an ongoing administrative and policy challenge rather than a threat to state sovereignty or critical infrastructure.

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

Threat characteristics

• Imminence: Long-term. The algorithm has been deployed since 2017 and represents an ongoing, systemic clinical and policy challenge rather than an immediate national security crisis.
• Autonomy: Human-supervised. The REACH VET algorithm automatically flags high-risk veterans, but human clinical staff make the final decision regarding outreach and treatment.
• Novelty: Established threat. Algorithmic bias and demographic disparities in machine learning models are well-documented and established challenges in AI deployment.

Impact by dimension

• Physical security: Minor. While not a kinetic attack or critical infrastructure failure, the algorithm's bias in suicide risk prediction poses indirect life-safety risks to a subset of the military veteran population.
• Information security: Negligible. No evidence of intelligence compromise, disinformation campaigns, or cyber espionage associated with this clinical algorithm.
• Sovereignty: Minor. Represents a failure in federal agency healthcare service delivery and resource allocation, but does not threaten state authority, territorial control, or core constitutional processes.
• Economic security: Negligible. No significant impact on financial systems, strategic industries, or overall national technological competitiveness.
• Societal stability: Minor. Involves systemic algorithmic discrimination and unequal treatment of female veterans in federal healthcare, but does not threaten overall social cohesion or trigger widespread civil unrest.