Argentinian City Government Deployed Teenage-Pregnancy Predictive Algorithm Using Invasive Demographic Data

In 2018, during the abortion-decriminalization debate in Argentina, the Salta city government deployed a teenage-pregnancy predictive algorithm built by Microsoft that allegedly lacked a defined purpose, explicitly considered sensitive information such as disability and whether their home had access to hot water.

Risk classification

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

The system targeted low-income, migrant, and Indigenous girls, unfairly labeling them as 'predestined' for teenage pregnancy based on biased socioeconomic variables, which represents unfair discrimination and misrepresentation.

Additional risk subdomains

• 2.1 Compromise of privacy by obtaining, leaking or correctly inferring sensitive information: The system relied on highly invasive data collection, including GPS locations, photos, and personal details of 12,000 young girls.
• 7.3 Lack of capability or robustness: The algorithm suffered from severe technical flaws, including data contamination and improper use of SMOTE, making its accuracy claims scientifically invalid.

Causal factors

• Entity: Human
• Intent: Intentional
• Timing: Post-deployment

The risk and resulting harms were caused by the intentional decision of human actors in the Salta government and Microsoft to deploy an invasive, unregulated predictive surveillance system targeting vulnerable populations.

EU AI Act risk tier

• Risk tier: 1 Unacceptable

Unacceptable Risk: The system is an exploitative tool targeting vulnerable populations based on age, disability, and socioeconomic status for state surveillance and population control, which violates fundamental rights.

AI system and alleged parties

• AI system: Technology Platform for Social Intervention (Microsoft)
• AI purpose: Behavioral Modeling; Child Welfare Risk Assessment
• Behaviour type: Tool
• Alleged developer: Microsoft
• Alleged deployer: Salta city government
• Alleged harmed parties: Salta teenage girls, Salta girls of minority groups

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 Substantial
• Democracy: direct Substantial, indirect Negligible
• Privacy: direct Substantial, indirect Substantial
• Psychological: direct Negligible, indirect Negligible
• Epistemic: direct Minor, indirect Substantial
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Differential treatment

Reported: The report explicitly describes systematic differential treatment of low-income, migrant, and Indigenous girls who were targeted by the system.

Directly caused: The AI system specifically targeted 12,000 girls and young women aged 10-19 from marginalized communities, labeling them as 'predestined' for teenage pregnancy based on socioeconomic variables.

Indirectly caused: This targeting reinforced systemic biases and eugenic-style population control, placing the blame for teenage pregnancy solely on vulnerable individuals while ignoring broader public health contexts.

Inferred additional harm: It is likely that the labeled girls faced social stigmatization and unequal treatment in their communities or schools due to being flagged by the system.

Civil rights

Reported: The report explicitly describes violations of the rights of girls and women, including their sexual and reproductive rights and right to be free from invasive surveillance.

Directly caused: Territorial agents subjected 12,000 girls to intimate surveillance, taking photos, recording GPS locations, and demanding sensitive personal information without clear consent or regulatory oversight.

Indirectly caused: The deployment of the system by the Ministry of Early Childhood created a coercive environment where vulnerable families could not refuse participation for fear of losing essential state benefits like vaccinations and free milk.

Democracy

Reported: The report describes the system as an extension of state authoritarianism and population control without regulatory oversight.

Directly caused: The provincial government deployed an untested, unregulated algorithmic surveillance system on its citizens without formal impact assessments or public accountability.

Privacy

Reported: The report explicitly describes invasive privacy violations of girls and women in Salta.

Directly caused: Territorial agents visited the homes of 12,000 girls, collecting highly sensitive personal data (including ethnicity, disability, and home infrastructure), taking photos, and recording GPS locations.

Indirectly caused: This sensitive data was fed into an opaque database managed by the government and Microsoft without clear data protection or privacy safeguards.

Epistemic

Reported: The report describes how the system generated misleading and scientifically invalid predictions about teenage pregnancy rates.

Directly caused: The algorithm produced flawed predictions claiming specific girls were '86 percent predestined' to become pregnant, which was presented as scientific truth on national television.

Indirectly caused: The dissemination of these false metrics misled policymakers and the public, potentially undermining evidence-based public health strategies.

People affected

• Occurrences reported: 1
• People reportedly harmed: 12000
• People reportedly exposed: 200000

Potential causes

Management

• Prioritizing PR over Safety: Officials promoted the tool on TV before verifying its technical validity.
• Lack of Accountability: Government and Microsoft failed to address criticisms or publish outcomes.
• Inadequate Risk Assessment: Management failed to evaluate ethical and social impacts on young girls.

Technology

• Data Contamination via SMOTE: SMOTE applied before data split artificially inflated the accuracy.
• Methodological Design Errors: Used current pregnancy status to predict future pregnancy years later.
• Opaque Algorithmic Models: The social theories and algorithms used in the system were completely opaque.

Data Inputs

• Incomplete and Biased Data: Sensitive data was hidden, leading to biased socioeconomic representations.
• Missing Critical Variables: Excluded key public health variables like sex education and contraception.
• Invasive Data Collection: Agents gathered photos, GPS locations, and sensitive surveys from minors.

Human Factors

• Overreliance on AI Predictions: Officials treated flawed predictions as absolute, predestined truths.
• Power Asymmetry and Coercion: Vulnerable families could not refuse due to dependency on state aid.
• Political and Ideological Bias: Anti-abortion groups used the tool to argue against reproductive rights.

Process and Methods

• No Independent Validation: The system lacked external technical validation before public deployment.
• Lack of Transparency: No official reports or data on accuracy were ever published.
• Poor Machine Learning Practices: Basic ML principles were violated by mixing training and evaluation data.

Regulatory Environment

• Absence of AI Regulation: No national laws existed to regulate or restrict AI system deployments.
• Lack of Oversight and Audits: No government body assessed the system's impact on girls and women.
• No Human Rights Framework: Lacked legal protections for vulnerable groups against algorithmic abuse.

Information quality

• Classification confidence: High
• Reason for confidence: The reports provide consistent, detailed accounts of the system's deployment, the specific data collected, the target demographic, and the technical flaws identified by the University of Buenos Aires. The technical analysis of the SMOTE data contamination is highly specific and credible.
• Ambiguities identified: It remains unclear whether the program was officially suspended or how the government ultimately utilized the collected data.
• Alternative interpretations: The developers framed the system as a pioneering public health tool to prevent teenage pregnancy, whereas critics interpreted it as an authoritarian surveillance and population control mechanism.

The Salta provincial government in Argentina deployed a highly flawed AI system developed with Microsoft to predict teenage pregnancy. The system targeted 12,000 marginalized and Indigenous girls, conducting invasive surveillance and violating fundamental privacy and human rights. While it did not threaten physical infrastructure or national sovereignty, it constitutes a substantial abuse of technology for state-sponsored algorithmic discrimination.

• Overall national security impact: Substantial
• Scope: Single nation
• Primary target: Argentina
• Alleged perpetrator: Salta Provincial Government

Threat characteristics

• Imminence: Long-term. The deployment represents a long-term policy and strategic concern regarding state surveillance rather than an active, imminent crisis.
• Autonomy: Human-controlled. The AI system functioned as a predictive tool to assist government agents, who maintained final decision-making control over interventions.
• Novelty: Evolved capability. Represents an evolution of predictive profiling technologies applied to sensitive public health and social intervention domains.

Impact by dimension

• Physical security: Negligible. The incident involved data collection and predictive modeling, with no kinetic impact or threat to critical infrastructure.
• Information security: Minor. While invasive data collection occurred, it did not involve state-level intelligence compromise or systematic foreign information warfare.
• Sovereignty: Minor. The deployment represents a provincial government program relying on a foreign tech partner, but did not threaten national sovereignty or core federal operations.
• Economic security: Negligible. The incident had minimal impact on national economic stability or strategic technological competitive advantage.
• Societal stability: Substantial. The system implemented invasive surveillance and algorithmic discrimination targeting 12,000 marginalized and Indigenous girls, violating fundamental human rights.