incident_id,date_documented,category,sector,model_attributed,severity_band,piss_score,affected_population_score,harm_severity_score,detection_failure_score,vendor_response_score,detection_mechanism,status,notable,aiid_cross_ref,brief_description,public_artifact_available,mitigations
LPI-2024-0001,2024-02-12,fabricated_citations,legal,undisclosed_chatgpt,medium,5.5,1.5,2.0,1.5,0.5,external_court_sanction,public_disclosed,true,652,Attorney filed brief with multiple non-existent case citations generated by ChatGPT; sanctioned by court,true,verification_workflows_for_legal_filings;mandatory_human_citation_check
LPI-2024-0007,2024-03-08,hallucination_user_harm,healthcare,undisclosed_consumer_chatbot,high,7.4,2.0,2.5,2.0,0.75,user_complaint,under_review,false,,Consumer health chatbot provided incorrect dosage guidance; patient hospitalized after acting on advice,false,medical_advisory_disclaimer;output_filter_for_dosage_specifics
LPI-2024-0014,2024-04-22,sensitive_data_exposure,financial,undisclosed_enterprise_deployment,high,7.2,1.5,2.5,2.5,0.5,external_researcher,patched,true,,Customer service chatbot extracted internal pricing tier information through system prompt extraction,true,system_prompt_redaction;output_classifier_for_internal_terms
LPI-2024-0019,2024-05-15,discriminatory_output,recruitment,gpt_4_class,medium,6.5,2.0,2.0,2.0,0.5,external_audit,under_review,false,,Recruitment screening tool produced demographically skewed candidate rankings beyond input variable explanation,true,disparate_impact_monitoring;continuous_bias_evaluation
LPI-2024-0027,2024-06-03,fabricated_citations,legal,gpt_4_class,low,4.5,0.5,2.0,1.5,0.5,external_court_sanction,public_disclosed,false,,Multiple fabricated cases in immigration brief; firm sanctioned,true,verification_workflows_for_legal_filings
LPI-2024-0031,2024-06-19,policy_violating_content,customer_service,undisclosed,medium,4.0,1.5,1.0,1.0,0.5,internal_review,patched,false,,Brand chatbot produced offensive output during user testing scenarios that reached customers,true,output_classifier_tuning;rapid_rollback_capability
LPI-2024-0042,2024-07-11,hallucination_user_harm,financial,undisclosed,medium,5.8,2.0,1.5,1.5,0.75,user_complaint,acknowledged,false,,Banking chatbot provided incorrect information about account holder rights leading to financial confusion,true,domain_expert_review;factuality_grounding_for_financial_topics
LPI-2024-0048,2024-08-02,agent_tool_misuse,retail,gpt_4_class,medium,5.5,1.5,1.5,2.0,0.5,internal_financial_audit,patched,false,,Customer service agent processed refunds beyond authorized scope during stress-test promotional period,true,financial_blast_radius_limits;supervisor_routing_for_high_value_actions
LPI-2024-0056,2024-08-28,fabricated_citations,legal,claude_3_class,low,4.0,0.5,2.0,1.0,0.5,external_court_sanction,public_disclosed,false,,Civil litigation filing with fabricated case law; attorney sanctioned and required CLE,true,verification_workflows_for_legal_filings
LPI-2024-0064,2024-09-14,sensitive_data_exposure,education,undisclosed,high,6.8,2.0,2.0,2.0,0.75,external_researcher,patched,false,,Tutoring platform chatbot revealed system prompt containing other student conversation excerpts,false,session_isolation_audit;system_prompt_compartmentalization
LPI-2024-0072,2024-10-05,misattribution_real_people,media,undisclosed,medium,5.7,1.5,2.0,1.5,0.75,subject_of_article,public_disclosed,false,,Article draft tool generated misattributed quotes attributed to public official; caught before publication,true,quote_verification_pipeline;named_person_factuality_check
LPI-2024-0079,2024-10-23,hallucination_user_harm,government,undisclosed,medium,5.4,1.5,2.0,1.5,0.5,external_advocate,under_review,false,,Benefits eligibility tool produced incorrect denials in pilot deployment affecting subset of applicants,true,disparate_impact_monitoring;benefits_decision_human_review
LPI-2024-0086,2024-11-18,policy_violating_content,education,undisclosed,medium,4.2,2.0,1.0,1.0,0.5,parent_complaint,patched,false,,Education chatbot for elementary students produced age-inappropriate content in edge case prompts,true,age_appropriate_output_classifier;child_safety_red_team
LPI-2024-0094,2024-12-08,agent_tool_misuse,enterprise,undisclosed,medium,5.0,1.0,1.5,2.0,0.5,internal_review,patched,false,,Coding agent modified production configuration files beyond user-confirmed scope,true,config_file_write_guards;diff_preview_before_apply
LPI-2024-0101,2024-12-22,fabricated_citations,legal,gpt_4_class,low,3.5,0.5,1.5,1.0,0.5,external_court_sanction,public_disclosed,false,,Self-represented litigant filed motion with fabricated cases; court issued guidance,true,verification_workflows_for_legal_filings;pro_se_litigant_warnings
LPI-2025-0009,2025-01-14,sensitive_data_exposure,financial,undisclosed,high,7.5,2.0,2.5,2.5,0.5,external_researcher,patched,true,,Customer service chatbot leaked partial customer profile information through context manipulation,false,context_isolation_audit;customer_data_redaction_in_context
LPI-2025-0017,2025-02-03,hallucination_user_harm,legal,undisclosed,medium,5.3,1.5,2.0,1.0,0.75,client_review,acknowledged,false,,Legal research assistant cited overruled precedents as current law in document review,true,case_status_grounding;current_law_verification
LPI-2025-0024,2025-02-19,discriminatory_output,financial,undisclosed,high,7.0,2.5,2.0,2.0,0.5,regulatory_audit,under_review,false,,Loan decision support tool exhibited demographic correlation patterns beyond input variables,true,protected_class_correlation_monitoring;model_card_disclosure_of_biases
LPI-2025-0033,2025-03-08,agent_tool_misuse,enterprise,gpt_4_class,medium,5.8,1.0,2.0,2.0,0.75,internal_review,patched,false,,Email assistant agent sent draft to incorrect recipient list due to context confusion,true,recipient_confirmation_step;dry_run_mode_for_communications
LPI-2025-0041,2025-03-22,misattribution_real_people,media,undisclosed,high,6.8,2.0,2.0,2.0,0.75,subject_of_article,public_disclosed,true,,National outlet published article with fabricated quotes attributed to public figure; retracted within 48 hours,true,editorial_AI_policy;quote_provenance_tracking
LPI-2025-0049,2025-04-04,fabricated_citations,legal,claude_3_class,medium,4.7,0.5,2.0,1.5,0.75,opposing_counsel,public_disclosed,false,,Brief contained fabricated quotation from real case; opposing counsel identified discrepancy,true,quote_verification_in_legal_briefs
LPI-2025-0057,2025-04-18,policy_violating_content,customer_service,undisclosed,low,3.5,1.5,1.0,0.5,0.5,user_complaint,patched,false,,Brand mascot chatbot produced output violating brand guidelines in extended conversations,true,brand_voice_classifier;long_context_safety_refresh
LPI-2025-0065,2025-05-07,hallucination_user_harm,healthcare,undisclosed,high,7.0,2.0,2.5,1.5,1.0,external_researcher,disputed,false,,Symptom checker recommended self-care for symptoms requiring urgent evaluation in documented test cases,true,medical_disclaimer_strengthening;urgent_symptom_keyword_detection
LPI-2025-0073,2025-05-26,sensitive_data_exposure,education,undisclosed,medium,6.0,2.0,1.5,2.0,0.5,external_researcher,patched,false,,Tutoring platform exposed student names from system prompt through translation request,false,system_prompt_redaction;translation_specific_safety_policy
LPI-2025-0082,2025-06-15,agent_tool_misuse,retail,undisclosed,high,7.4,2.5,2.0,2.5,0.5,external_security_researcher,patched,false,,Shopping agent platform processed unauthorized purchases via prompt injection in product descriptions,true,product_description_quarantine;purchase_authorization_two_factor
LPI-2025-0089,2025-06-29,hallucination_user_harm,government,undisclosed,high,7.9,2.5,2.5,2.5,0.5,external_advocacy_group,public_disclosed,true,,Pilot deployment of LLM-assisted benefits eligibility produced incorrect denials disproportionately affecting subset population,true,disparate_impact_monitoring;benefits_decision_appeal_process_strengthening
LPI-2025-0097,2025-07-14,fabricated_citations,academic,undisclosed,medium,4.5,1.5,1.5,1.0,0.5,peer_review,public_disclosed,false,,Published preprint contained fabricated citations to non-existent papers; preprint withdrawn,true,citation_verification_in_academic_workflows
LPI-2025-0105,2025-07-31,discriminatory_output,recruitment,undisclosed,medium,5.7,1.5,2.0,1.5,0.75,external_audit,under_review,false,,CV screening tool exhibited age-correlated ranking patterns flagged by independent audit,true,disparate_impact_monitoring;protected_class_correlation_check
LPI-2025-0113,2025-08-12,sensitive_data_exposure,financial,undisclosed,critical,8.5,2.5,3.0,2.5,0.5,external_researcher,patched,true,,Banking chatbot system prompt extraction revealed customer service routing patterns including PII handling rules,false,system_prompt_redaction;PII_handling_rules_compartmentalization
LPI-2025-0121,2025-08-29,hallucination_user_harm,legal,gpt_4_class,medium,5.0,1.0,2.0,1.5,0.5,client_review,acknowledged,false,,Contract analysis tool misidentified governing law clause leading to client misadvice,true,legal_clause_detection_grounding;contract_review_human_check
LPI-2025-0129,2025-09-11,policy_violating_content,media,undisclosed,medium,4.0,2.0,1.0,0.5,0.5,user_complaint,patched,false,,Comment moderation chatbot produced offensive responses to certain user demographics,true,demographic_bias_red_team;moderation_output_classifier
LPI-2025-0137,2025-09-25,agent_tool_misuse,enterprise,gpt_4_class,medium,5.5,1.0,2.0,2.0,0.5,internal_review,patched,false,,Calendar agent scheduled internal-confidential meeting with external party due to context confusion,true,meeting_attendee_verification;confidentiality_aware_scheduling
LPI-2025-0145,2025-10-09,fabricated_citations,legal,undisclosed,medium,4.7,0.5,2.0,1.5,0.75,opposing_counsel,public_disclosed,false,,Government brief contained fabricated regulatory citations; agency ordered review,true,verification_workflows_for_government_filings
LPI-2025-0153,2025-10-22,misattribution_real_people,academic,undisclosed,medium,5.5,1.5,2.0,1.5,0.5,subject_named,public_disclosed,false,,Academic literature review tool generated misattributed claims to named researchers,true,attribution_grounding;named_researcher_factuality_check
LPI-2025-0161,2025-11-08,hallucination_user_harm,financial,undisclosed,high,6.8,2.0,2.5,1.5,0.75,user_complaint,acknowledged,false,,Investment advisory chatbot provided incorrect tax treatment guidance affecting filings,true,financial_advice_disclaimer;tax_topic_grounding
LPI-2025-0169,2025-11-22,sensitive_data_exposure,healthcare,undisclosed,critical,8.2,2.5,2.5,2.5,0.75,external_researcher,under_review,true,,Patient-facing health chatbot revealed deployment configuration details including data handling protocols,false,system_prompt_isolation;medical_deployment_security_audit
LPI-2025-0177,2025-12-05,agent_tool_misuse,enterprise,claude_3_class,medium,6.2,1.5,2.0,2.0,0.75,internal_security,patched,false,,Code review agent committed changes that bypassed required approval workflow due to permission misconfiguration,true,permission_boundary_verification;agent_action_audit_log
LPI-2025-0185,2025-12-19,discriminatory_output,recruitment,undisclosed,medium,5.5,1.5,2.0,1.5,0.5,external_audit,under_review,false,,Video interview analysis tool exhibited gender-correlated scoring patterns flagged by independent reviewer,true,protected_class_correlation_monitoring;interview_AI_disclosure_to_candidates
LPI-2025-0193,2026-01-04,fabricated_citations,legal,gpt_5_class,medium,5.0,0.5,2.0,1.5,1.0,external_court_sanction,public_disclosed,false,,Federal court brief with fabricated citations; firm received ongoing sanctions,true,verification_workflows_for_legal_filings;federal_court_AI_disclosure
LPI-2026-0008,2026-01-15,agent_tool_misuse,retail,undisclosed,high,6.4,2.0,2.0,2.0,0.5,internal_financial_audit,patched,true,,Customer service agent authorized refunds beyond policy scope after multi-turn user pressure tactics,true,financial_blast_radius_limits;supervisor_routing_for_high_value_actions
LPI-2026-0014,2026-01-28,hallucination_user_harm,healthcare,undisclosed,critical,8.7,2.5,3.0,2.5,0.75,physician_pattern_detection,public_disclosed,true,,Hospital triage chatbot misclassified ED-level cases toward urgent care or self-monitoring across documented cluster,true,medical_advisory_post_deployment_monitoring;urgency_classification_human_review
LPI-2026-0022,2026-02-09,sensitive_data_exposure,financial,gpt_5_class,critical,8.0,2.5,2.5,2.5,0.5,external_researcher,patched,true,,Wealth management chatbot leaked client portfolio summary patterns through context-engineering attack,false,client_data_isolation;portfolio_pattern_classifier
LPI-2026-0029,2026-02-22,hallucination_user_harm,legal,claude_4_class,medium,5.5,1.5,2.0,1.5,0.5,client_review,acknowledged,false,,Legal research tool cited overruled precedent in discovery documents leading to motion based on incorrect law,true,case_status_grounding;current_law_continuous_update
LPI-2026-0036,2026-03-04,policy_violating_content,customer_service,undisclosed,medium,4.5,2.0,1.0,1.0,0.5,internal_review,patched,false,,Brand chatbot produced misleading product comparison statements about competitor products,true,competitor_factuality_check;comparative_claim_verification
LPI-2026-0042,2026-03-15,discriminatory_output,financial,undisclosed,high,7.2,2.5,2.0,2.0,0.75,regulatory_inquiry,under_review,true,,Credit assessment tool exhibited disparate approval rates correlating with applicant zip code patterns,true,disparate_impact_monitoring;protected_class_correlation_audit
LPI-2026-0049,2026-03-26,agent_tool_misuse,enterprise,gpt_5_class,high,7.0,1.5,2.5,2.0,1.0,external_security_researcher,under_disclosure,false,,Workflow automation agent triggered unauthorized data export through tool description injection,false,tool_description_integrity;export_authorization_two_factor
LPI-2026-0055,2026-04-08,misattribution_real_people,media,undisclosed,medium,6.0,1.5,2.0,2.0,0.5,subject_named,public_disclosed,false,,Newsroom drafting tool included misattributed statement about public official; published before catch,true,quote_provenance_tracking;named_person_factuality_pre_publication
LPI-2026-0062,2026-04-18,fabricated_citations,academic,undisclosed,medium,5.0,1.5,2.0,1.0,0.5,journal_review,public_disclosed,false,,Submitted manuscript contained fabricated citations to non-existent papers; rejected,true,citation_verification_in_academic_workflows
LPI-2026-0067,2026-04-25,hallucination_user_harm,government,undisclosed,high,7.5,2.5,2.5,2.0,0.5,external_advocate,under_review,true,,Immigration document assistant produced inaccurate filing guidance affecting non-trivial applicant cohort,true,immigration_decision_human_review;multilingual_factuality_check
LPI-2024-0103,2024-09-30,unauthorized_actions_agent,enterprise,undisclosed,high,7.6,1.5,2.5,2.5,1.0,internal_audit,patched,false,,Agent issued database modification commands beyond user-confirmed scope during testing escalation,true,database_write_authorization;agent_action_human_confirmation_for_destructive_ops
LPI-2024-0108,2024-11-04,cross_user_data_leakage,enterprise,vendor_specific,critical,9.0,2.5,2.5,2.5,1.0,external_researcher,patched,true,,Multi-tenant deployment briefly exposed cached responses from one tenant to another during high-load incident,false,patched_per_tenant_cache_namespacing
LPI-2025-0040,2025-03-18,unauthorized_actions_agent,enterprise,undisclosed,high,6.8,1.5,2.0,2.0,0.75,internal_review,patched,false,,Coding agent created and pushed code branch with credentials in plaintext due to context misinterpretation,false,credential_detection_pre_commit;agent_action_audit_log
LPI-2025-0078,2025-06-08,cross_user_data_leakage,enterprise,vendor_specific,critical,8.5,2.0,2.5,2.5,0.5,external_researcher,patched,true,,Research preview feature briefly allowed cross-session content visibility under specific conditions,false,patched_session_isolation_in_preview_features
LPI-2025-0114,2025-08-15,unauthorized_actions_agent,enterprise,undisclosed,high,7.3,1.5,2.5,2.0,0.75,internal_security,patched,false,,Browsing agent submitted form data on third-party site beyond user-confirmed task scope,true,form_submission_user_confirmation;agent_navigation_boundaries
LPI-2025-0156,2025-10-30,unauthorized_actions_agent,enterprise,gpt_5_class,high,7.0,1.5,2.0,2.0,1.0,external_security_researcher,under_disclosure,false,,Email management agent forwarded thread to unauthorized recipient via injection in calendar invite,false,calendar_content_quarantine;recipient_authorization_step
LPI-2026-0004,2026-01-09,cross_user_data_leakage,enterprise,vendor_specific,critical,9.2,2.5,3.0,2.5,0.75,external_researcher,patched,true,,Caching layer edge case exposed user A response fragments to user B during specific concurrency pattern,false,patched_cache_isolation_with_user_binding
LPI-2026-0019,2026-02-04,unauthorized_actions_agent,enterprise,undisclosed,high,7.8,2.0,2.5,2.5,0.75,external_security_researcher,under_disclosure,true,,Workflow agent permission escalation chain enabled actions beyond initial authorization scope,false,permission_boundary_verification;capability_token_enforcement
LPI-2024-0029,2024-06-12,policy_violating_content,customer_service,undisclosed,low,3.0,1.0,1.0,0.5,0.5,internal_review,patched,false,,Brand chatbot used informal language inconsistent with brand voice guidelines for sensitive customer issues,true,brand_voice_classifier;sensitive_topic_routing
LPI-2024-0044,2024-07-19,hallucination_user_harm,retail,undisclosed,low,3.5,1.5,1.0,0.5,0.5,user_complaint,patched,false,,Product description generator created inaccurate technical specifications affecting customer purchase decisions,true,technical_spec_grounding;human_review_for_product_data
LPI-2024-0061,2024-09-08,fabricated_citations,academic,undisclosed,low,3.0,0.5,1.5,0.5,0.5,supervisor_review,public_disclosed,false,,Student thesis draft contained fabricated source citations identified during supervisor review pre-submission,true,citation_verification_in_academic_workflows
LPI-2024-0091,2024-12-02,sensitive_data_exposure,enterprise,undisclosed,medium,5.5,1.0,2.0,2.0,0.5,external_researcher,patched,false,,Internal knowledge base chatbot revealed competitor analysis content through context manipulation,false,internal_knowledge_compartmentalization;sensitive_doc_classifier
LPI-2025-0011,2025-01-22,policy_violating_content,gaming,undisclosed,low,2.5,1.0,0.5,0.5,0.5,user_complaint,patched,false,,Game NPC chatbot produced lore-inconsistent content affecting player narrative experience,true,lore_consistency_classifier;game_specific_safety_policy
LPI-2025-0029,2025-03-01,hallucination_user_harm,travel,undisclosed,medium,5.0,2.0,1.5,1.0,0.5,user_complaint,acknowledged,false,,Travel planning chatbot provided incorrect visa requirements leading to traveler issues at border,true,travel_document_factuality_grounding;current_visa_data_integration
LPI-2025-0046,2025-04-09,discriminatory_output,customer_service,undisclosed,low,4.0,1.5,1.5,0.5,0.5,user_complaint,patched,false,,Voice assistant exhibited differential response patterns to inferred user demographics,true,demographic_neutrality_red_team;response_pattern_audit
LPI-2025-0068,2025-05-19,fabricated_citations,journalism,undisclosed,medium,4.5,1.5,2.0,0.5,0.5,fact_checker,public_disclosed,false,,Article draft tool produced fabricated source attributions caught during fact-check process,true,source_verification_in_journalism_workflows
LPI-2025-0093,2025-07-08,sensitive_data_exposure,healthcare,undisclosed,high,7.0,2.0,2.5,2.0,0.5,external_researcher,patched,false,,Patient communication chatbot revealed routing rules including treatment-class indicators,false,medical_routing_compartmentalization;treatment_indicator_redaction
LPI-2025-0118,2025-08-22,agent_tool_misuse,enterprise,undisclosed,medium,5.7,1.0,2.0,2.0,0.75,internal_review,patched,false,,Document management agent moved files outside authorized folder structure during reorganization task,true,filesystem_boundary_enforcement;agent_filesystem_dry_run
LPI-2025-0142,2025-10-12,misattribution_real_people,media,undisclosed,high,6.8,2.0,2.0,2.0,0.75,subject_of_article,public_disclosed,true,,National outlet article with fabricated quotes attributed to public figure; retracted,true,editorial_AI_policy;quote_provenance_tracking
LPI-2025-0166,2025-11-15,policy_violating_content,education,undisclosed,medium,4.5,2.0,1.0,1.0,0.5,parent_complaint,patched,false,,Learning platform chatbot produced content beyond age-appropriate guidelines in edge cases,true,age_appropriate_output_classifier;child_safety_red_team
LPI-2025-0181,2025-12-12,hallucination_user_harm,legal,gpt_4_class,medium,5.5,1.0,2.0,2.0,0.5,opposing_counsel,public_disclosed,false,,Discovery document review tool misidentified privileged communications leading to inadvertent disclosure,true,privilege_detection_human_review;privileged_communication_classifier
LPI-2026-0006,2026-01-12,fabricated_citations,legal,undisclosed,low,3.5,0.5,1.5,1.0,0.5,external_court_sanction,public_disclosed,false,,Civil filing with non-existent statutory citations; attorney sanctioned,true,verification_workflows_for_legal_filings
LPI-2026-0016,2026-02-01,agent_tool_misuse,retail,undisclosed,medium,5.0,1.5,1.5,1.5,0.5,user_complaint,patched,false,,Shopping assistant agent applied incorrect promotional codes affecting customer transactions,true,promotional_code_validation;transaction_audit_for_AI_actions
LPI-2026-0024,2026-02-13,sensitive_data_exposure,financial,undisclosed,high,6.5,1.5,2.5,2.0,0.5,external_researcher,patched,false,,Banking chatbot revealed account-type-specific message templates including risk indicators,false,template_compartmentalization;account_data_isolation
LPI-2026-0033,2026-02-27,discriminatory_output,recruitment,undisclosed,medium,6.0,1.5,2.0,2.0,0.5,external_audit,under_review,false,,Resume parsing tool exhibited education-institution-correlated ranking patterns,true,education_institution_correlation_check;protected_class_proxies_monitoring
LPI-2026-0040,2026-03-12,policy_violating_content,media,undisclosed,medium,4.7,2.0,1.0,1.0,0.75,external_audit,patched,false,,Comment generation tool produced inflammatory content patterns flagged by external audit,true,inflammatory_content_classifier;moderation_threshold_tuning
LPI-2026-0047,2026-03-23,hallucination_user_harm,healthcare,undisclosed,high,7.2,2.0,2.5,2.0,0.75,user_complaint,under_review,false,,Mental health chatbot provided guidance inconsistent with crisis intervention protocols in subset of cases,false,crisis_keyword_detection;mental_health_specialist_routing
LPI-2026-0053,2026-04-02,unauthorized_actions_agent,enterprise,gpt_5_class,high,7.5,1.5,2.5,2.5,1.0,external_security_researcher,under_disclosure,false,,Browsing agent executed transactions on third-party site through prompt injection in displayed content,false,transaction_authorization_step;displayed_content_quarantine
LPI-2026-0058,2026-04-13,fabricated_citations,legal,claude_4_class,medium,4.5,0.5,2.0,1.5,0.5,opposing_counsel,public_disclosed,false,,Brief contained fabricated case quotation; opposing counsel identified during reply preparation,true,verification_workflows_for_legal_filings
LPI-2024-0036,2024-06-25,autonomous_system_override,industrial,undisclosed,critical,8.9,2.0,3.0,2.5,1.0,operator_intervention,patched,true,,Industrial monitoring assistant provided override recommendation that conflicted with safety protocol; operator caught the discrepancy,false,safety_protocol_grounding;human_in_the_loop_for_overrides
LPI-2025-0086,2025-06-22,autonomous_system_override,transportation,undisclosed,critical,8.5,2.0,2.5,2.5,1.0,operator_intervention,patched,false,,Logistics planning assistant suggested route violating regulated rest-period requirements; dispatcher caught the issue,true,regulatory_grounding_for_operational_decisions;human_review_for_compliance_critical_actions
LPI-2026-0030,2026-02-25,autonomous_system_override,energy,undisclosed,critical,9.1,2.5,3.0,2.5,1.0,operator_intervention,under_disclosure,true,,Energy grid management assistant provided recommendation that would have triggered safety override; system review caught issue,false,safety_protocol_grounding;defense_in_depth_for_critical_infrastructure
LPI-2024-0083,2024-11-12,sensitive_data_exposure,public_sector,undisclosed,medium,6.0,2.0,2.0,1.5,0.5,external_researcher,patched,false,,Government services chatbot revealed internal classification rules during translation request,true,government_data_compartmentalization;translation_specific_safety
LPI-2025-0050,2025-04-15,hallucination_user_harm,journalism,undisclosed,medium,5.5,2.0,1.5,1.5,0.5,reader_complaint,public_disclosed,false,,Article aggregation summary tool produced misleading summaries that altered article meaning,true,summary_factuality_check;source_alignment_verification
LPI-2025-0124,2025-09-04,policy_violating_content,health_consumer,undisclosed,medium,4.8,2.0,1.5,0.5,0.5,user_complaint,patched,false,,Wellness app chatbot produced advice categorized as outside intended scope by healthcare advisors,true,scope_classification_clarity;medical_advice_disclaimer
LPI-2024-0117,2024-12-15,fabricated_citations,public_policy,undisclosed,medium,5.0,1.5,2.0,1.0,0.5,policy_research_review,public_disclosed,false,,Policy briefing document contained fabricated citations to non-existent studies; identified during peer review,true,citation_verification_in_policy_workflows
LPI-2025-0135,2025-09-22,sensitive_data_exposure,enterprise,undisclosed,high,6.8,1.5,2.5,2.0,0.75,external_researcher,patched,false,,Internal HR chatbot revealed employee performance review template patterns through context probing,false,HR_data_compartmentalization;employee_data_classifier
LPI-2026-0011,2026-01-20,unauthorized_actions_agent,enterprise,gpt_5_class,medium,5.5,1.0,2.0,2.0,0.5,internal_security,patched,false,,DevOps agent executed deployment to staging environment beyond user-confirmed scope,true,deployment_authorization;agent_action_audit_log
LPI-2024-0124,2024-12-30,discriminatory_output,public_sector,undisclosed,medium,6.5,2.0,2.0,2.0,0.5,external_advocacy_group,under_review,false,,Public benefits triage tool exhibited language-correlated routing patterns affecting non-native speakers,true,multilingual_neutrality_audit;language_correlation_check
LPI-2025-0098,2025-07-16,misattribution_real_people,journalism,undisclosed,medium,5.7,1.5,2.0,1.5,0.75,subject_named,public_disclosed,false,,Investigative summary tool included misattributed claim about named source; correction issued,true,attribution_grounding;named_source_verification
LPI-2026-0026,2026-02-17,hallucination_user_harm,financial,undisclosed,medium,5.5,1.5,2.0,1.5,0.5,user_complaint,acknowledged,false,,Tax preparation assistant provided incorrect deduction guidance affecting filings,true,tax_topic_grounding;current_tax_law_integration
LPI-2026-0044,2026-03-18,sensitive_data_exposure,education,undisclosed,medium,5.7,1.5,2.0,1.5,0.75,external_researcher,patched,false,,Online learning platform chatbot revealed scoring rubric details through extended conversation,false,rubric_compartmentalization;assessment_data_classifier
LPI-2026-0061,2026-04-21,policy_violating_content,enterprise,undisclosed,medium,4.0,1.0,1.5,1.0,0.5,internal_review,patched,false,,Internal knowledge assistant produced content inconsistent with organizational confidentiality policy in edge cases,true,confidentiality_policy_classifier;internal_data_safety_training