{"schema_version":"0.10.0","id":"LR-2026-0546","record_version":1,"title":"Kenya Social Health Authority means testing algorithm overcharges poor households and undercharges the wealthy","verification":{"human_verified":false,"method":"automated_gate","auto_verified":true,"score":44.5,"score_breakdown":{"source_authority":12.1,"independent_corroboration":7.0,"implicated_party_confirmed":0.0,"citation_integrity":14.0,"specificity":0.0,"completeness":4.48,"unhedged_language":4.71},"checks":{"schema_valid":true,"every_field_cited":true,"quotes_verbatim":true,"source_fetched":true,"distinct_incident":true,"dates_plausible":true,"sources_agree":true,"source_describes_incident":true,"severe_backed_by_source":true,"record_integrity":true,"eligible":true,"about_this_case":true},"gate_version":"gate-2026-10-18.7","cross_check_model":null,"cross_check_agreement":null,"duplicate_of":null,"verified_by":null,"verified_at":"2026-11-01","extraction_run_id":null,"evidence_level":{"level":"preliminary","reason":"Publisher not yet named · subject to change","original_document":null,"reports_total":1,"reports_independent":0,"reports_derivative":0,"developing":false,"disputed":false,"rules_version":"evidence-2026-10-29.1","computed_at":"2026-11-01","history":[{"date":"2026-11-01","from_level":null,"to_level":"preliminary","reason":"Publisher not yet named · subject to change"}]},"sources_hash":"b1a70b7ad131f788c98aa0dbe6d904ac94d6b5d356c70c22741088a798d56326"},"system":{"model_name":null,"model_version":null,"developer":"unknown","harness":null,"tools_granted":[],"permissions_scope":null,"autonomy_level":"unknown","guardrails_present":[],"deployment_context":"production","tier":null,"lifecycle_stage":"production","integration_type":null,"intended_use":"Assess how much money Kenyans should contribute in order to access the public healthcare system","safeguard_state":null,"production_intended":null},"event":{"date_occurred":null,"date_disclosed":"2026-05-04","incident_class":"bias_disparate_impact","failure_class":null,"agent_involved":null,"secondary_failure_classes":[],"trigger":"unknown","duration":null,"detection_method":"unknown","summary":"Kenya's Social Health Authority deployed a machine-learning means testing model, a linear regression trained on the 2021 Kenyan Continuous Households Survey, to set health insurance contributions for informal-sector workers. Under the reforms, workers pay 2.75 percent of estimated income. Lighthouse Reports, with Africa Uncensored and The Guardian, reconstructed the model from released training data, variables and formula. Their analysis found it overcharged the poorest households and undercharged the wealthiest. Among the poorest households, 80 percent had consumption overpredicted. An IDinsight evaluation before the 2024 launch had found the model inequitable, particularly for low-income households. The Social Health Authority did not respond to the detailed findings. The investigators note that their reconstruction may not match the deployed model exactly.","tags":[],"date_detected":null,"near_miss":null,"root_cause":null,"root_cause_detail":"Linear regression-based means testing tends to overestimate the incomes of the poorest households and understate those of the richest, a pattern rooted in the method's underlying mathematics","failure_class_detail":null,"responsible_party":"deployer","adversarial_actor":null,"input_provenance":null,"time_to_contain":null,"operations_paused":null,"remediation":null,"precursor_events":null,"primary_risk":"RK16"},"impact":{"direct_loss_usd":null,"remediation_hours":null,"records_exposed":null,"third_parties_affected_count":null,"severity":"major","severity_because":"Test: many third parties (a stated count of two or more). Fact: the model has calculated the public healthcare premiums of millions of Kenyans.","harm_class":["financial","discriminatory"],"affected_description":"Millions of Kenyans whose public healthcare premiums were calculated by the model","protected_class_involved":null,"data_categories_exposed":[],"loss_range_usd":null,"reversibility":null,"downstream_exposure_count":null},"evidence":{"sources":[{"url":"https://www.lighthousereports.com/methodology/how-we-investigated-kenyas-ai-means-testing-system/","publisher":"Lighthouse Reports","date":"2026-05-04","type":"press","title":"How we investigated Kenya’s AI Means Testing System - Lighthouse Reports","archived_url":null,"attachment":"first_report"}],"per_field_confidence":{"system.deployment_context":"C","system.lifecycle_stage":"C","system.intended_use":"C","event.root_cause_detail":"C","event.responsible_party":"C","impact.harm_class":"C","impact.affected_description":"C","event.incident_class":"D","impact.severity":"D","impact.severity_because":"D","event.primary_risk":"D","event.summary":"C","system.autonomy_level":"D","event.trigger":"D","event.detection_method":"D","context.country":"D","system.developer":"D","event.date_disclosed":"D","context.denominator_available":"D"},"field_citations":{"system.deployment_context":[{"source_index":0,"quote":"The algorithm, deployed by Kenya’s Social Health Authority, has calculated the public healthcare premiums of millions of Kenyans."}],"system.lifecycle_stage":[{"source_index":0,"quote":"The algorithm, deployed by Kenya’s Social Health Authority, has calculated the public healthcare premiums of millions of Kenyans."}],"system.intended_use":[{"source_index":0,"quote":"A central component was the development of an algorithm to assess how much money Kenyans should contribute in order to access the public healthcare system."}],"event.root_cause_detail":[{"source_index":0,"quote":"Linear regression–based means testing systems tend to overestimate the incomes of the poorest households while understating those of the richest, a pattern rooted in the method’s underlying mathematics."}],"event.responsible_party":[{"source_index":0,"quote":"The algorithm, deployed by Kenya’s Social Health Authority, has calculated the public healthcare premiums of millions of Kenyans."}],"impact.harm_class":[{"source_index":0,"quote":"In other words, the system overcharges the poorest Kenyans while undercharging the wealthiest Kenyans."}],"impact.affected_description":[{"source_index":0,"quote":"The algorithm, deployed by Kenya’s Social Health Authority, has calculated the public healthcare premiums of millions of Kenyans."}],"event.incident_class":[{"source_index":0,"quote":"The report, authored just before the system’s 2024 launch, found that the model was “inequitable, particularly for low-income households” and the 2021 Kenyan Continuous Households Survey used to train the model was out-of-date with the current socio-economic condition.”"}],"impact.severity":[{"source_index":0,"quote":"The algorithm, deployed by Kenya’s Social Health Authority, has calculated the public healthcare premiums of millions of Kenyans."}],"impact.severity_because":[{"source_index":0,"quote":"The algorithm, deployed by Kenya’s Social Health Authority, has calculated the public healthcare premiums of millions of Kenyans."}],"event.summary":[{"source_index":0,"quote":"The algorithm, deployed by Kenya’s Social Health Authority, has calculated the public healthcare premiums of millions of Kenyans."},{"source_index":0,"quote":"Under the new SHA reforms, informal sector workers are required to enrol and pay 2.75 percent of their income as a health insurance contribution."},{"source_index":0,"quote":"Amongst the poorest households, 16 percent are predicted correctly and 80 percent have their consumption overpredicted."},{"source_index":0,"quote":"In other words, the system overcharges the poorest Kenyans while undercharging the wealthiest Kenyans."},{"source_index":0,"quote":"The report, authored just before the system’s 2024 launch, found that the model was “inequitable, particularly for low-income households” and the 2021 Kenyan Continuous Households Survey used to train the model was out-of-date with the current socio-economic condition.”"},{"source_index":0,"quote":"The Social Health Authority did not respond to a detailed set of questions regarding the findings presented in this methodology."},{"source_index":0,"quote":"It’s possible that the agency may have carried out additional steps not reflected in the materials, such as data cleaning, imputation of missing values, or post-estimation calibration."}]},"notes_on_gaps":"The sources are a single press investigation by Lighthouse Reports. This is not an autonomous agent case: agent_involved is set false and no failure_class is given. The developer, model version, harness, tools and autonomy level are not stated. No occurrence date is stated for the harm itself; the October 2024 registration is cited only as an example and not used as date_occurred. No date of detection or direct loss, record count or third-party count is stated; 'millions of Kenyans' is used only for severity and affected description. Severity major is my judgement under the 'many third parties (stated count of two or more)' test. Root cause 'data' is my judgement from the IDinsight finding that the training survey was out-of-date; the investigation also attributes the bias to the inherent linear regression method, which is recorded in root_cause_detail. Responsible party 'deployer' is my judgement from the Social Health Authority deploying the model. Control I2 'failed' is my judgement: an evaluation was done but the model launched despite its findings; the sources do not say the evaluation was a realistic behaviour evaluation. The IDinsight recommendations were not confirmed as adopted; SHA's later FOI account differed, and no fix was reported, so remediations is empty. The reconstruction was made by the investigators and may not match the deployed model; the SHA did not respond to the findings. No other controls are stated as held, failed or missing, so none are listed. [case created by the backfill: read by claude-haiku-5-5, backfill-2026-11-01.2, 2026-11-01; not yet reviewed] [2026-11-01: evidence level set: Preliminary (Publisher not yet named · subject to change)]"},"context":{"industry":null,"org_size_band":null,"country":"unknown","denominator_available":false,"denominator":null,"related_incident_ids":[],"systemic_exposure":null,"shared_model_id":null,"upstream_incident_id":null,"incident_evidence":null,"regulatory":null,"coverage":null},"analysis":{"sections":{"timeline":[{"text":"Africa Uncensored sends a public access request for the model formula, variables and weights.","quote":"In August 2025, Africa Uncensored sent a public access request to the Social Health Authority requesting, among other things, the exact formula, variables and weights used in the model.","source_index":0,"when":"2025-08"},{"text":"The Social Health Authority responds, confirming training on the 2021 KCHS and disclosing the formula and variables but not the coefficients.","quote":"In November 2025, the Social Health Authority responded to our request, confirming that the model had trained on the 2021 KCHS and disclosing the model formula and the variables used, but not the coefficients for each variable.","source_index":0,"when":"2025-11"}]},"model":"claude-haiku-5-5","prompt_version":"backfill-2026-11-01.2","extracted_at":"2026-11-01","source_urls":["https://www.lighthousereports.com/methodology/how-we-investigated-kenyas-ai-means-testing-system/"]}}