OverviewRisksControlsMapReference

Search the Atlas

Search risks, controls, and glossary terms

CriticalHallucination●Realized

Hallucination / confabulation

Model Accuracy & Reliability

Description

The model generates confidently stated but false or fabricated content, presented with the same fluency as accurate output.

Example scenario

A customer-facing bot invents a non-existent tax-saving clause, exposing the bank to mis-selling claims.

Real-world evidence●Realized

AI hallucination producing factually incorrect outputs in production is among the most extensively documented real-world AI failure modes, confirmed in legal filings, regulatory proceedings, and journalism across multiple industries. The Mata v. Avianca court record (2023) alone documents fabricated citations presented as real in live legal proceedings.

Mata v. Avianca — AI-hallucinated case citations submitted to federal courtUS District Court, Southern District of New York · 2023Air Canada chatbot gave wrong information about bereavement faresBBC News · 2024ChatGPT is making things up about real people — and it's hard to fixBloomberg · 2023
Tested by FinProof
Document hallucination
Benchmarked by FinProof

Primary mitigations

  • RAG grounding with citations
  • abstention/uncertainty thresholds
  • output verification
  • human review for material decisions
  • constrained generation.

Detection signals

Groundedness/faithfulness scoring; claim-verification sampling; hallucination-rate dashboards.

Mitigating controls

21
Non-agentic controls
ZYC-ACCU-001Non-Agentic
Lexical Similarity
Accuracy
ZYC-XAI-001Non-Agentic
Field Justification
Explainability (XAI)
ZYC-ROB-001Non-Agentic
Out-of-Distribution Detection
Robustness
ZYC-XAI-005Non-Agentic
Source Citation
Explainability (XAI)
ZYC-ROB-011Non-Agentic
Abstention
Robustness
ZYC-ROB-012Non-Agentic
Extraction Hallucination
Robustness
ZYC-ACCU-016Non-Agentic
Semantic Label Accuracy
Accuracy
ZYC-ACCU-003Non-Agentic
Lexical Similarity
Accuracy
ZYC-ACCU-002Non-Agentic
Lexical Similarity
Accuracy
ZYC-ACCU-004Non-Agentic
Semantic Similarity
Accuracy
ZYC-ACCU-005Non-Agentic
Factuality
Accuracy
ZYC-ACCU-006Non-Agentic
Factuality
Accuracy
ZYC-ACCU-007Non-Agentic
Text Presence
Accuracy
ZYC-ACCU-008Non-Agentic
Field Detection Accuracy
Accuracy
ZYC-ACCU-009Non-Agentic
Field Extraction Accuracy
Accuracy
ZYC-ACCU-010Non-Agentic
Label/Entity Mapping Accuracy
Accuracy
ZYC-ROB-005Non-Agentic
OCR Noise Tolerance
Robustness
ZYC-ROB-006Non-Agentic
Template Generalization
Robustness
ZYC-ROB-007Non-Agentic
Non‑standard Field Success Rate
Robustness
ZYC-ROB-008Non-Agentic
Structural Awareness
Robustness
ZYC-OREL-002Non-Agentic
Factuality
Over-reliance

Related risks in Model Accuracy & Reliability

ZNR-MA-002High
Fabricated citations & references
ZNR-MA-003High
Overconfidence & poor calibration
ZNR-MA-004Medium
Quantitative & arithmetic reasoning errors
ZNR-MA-005Medium
Inconsistency / non-determinism