license: cc-by-4.0
task_categories:
- text-classification
- question-answering
language: - en
tags: - ai-safety
- ai-failures
- llm
- red-teaming
- risk-intelligence
- autonomous-agents
- alignment
size_categories: - n<1K
AI Failure Intelligence Dataset
Structured, analyst-grade dataset of real-world AI failures.
Built for researchers, red-teamers, and enterprise AI security teams.
Dataset Description
This is a FREE sample of 50 curated AI failure cases from the full
AI Failure Intelligence dataset (5,000+ cases, updated daily).
Each case is enriched with machine intelligence including:
- Failure classification across 14 categories
- Severity scoring (0-100)
- Root cause analysis
- Risk pattern identification
- Actionable recommendations
- Transferable lessons
Failure Type Categories
| Category | Description |
|---|---|
| Autonomous Agent Risk | AI agents taking unintended actions |
| Alignment Failure | Safety bypass, jailbreaks, value misalignment |
| Robustness Failure | Adversarial attacks, out-of-distribution failures |
| Benchmark/Evaluation Issue | Test contamination, metric gaming |
| Security | Exploits, CVEs, RCE vulnerabilities |
| Hallucination | Fabricated or false model outputs |
| Bias | Discriminatory or unfair outputs |
| Prompt Injection | Malicious input hijacking model behavior |
Dataset Structure
Title - Case title
Industry - Affected industry
Failure Type - Classification category
Severity Score - 0-100 severity rating
Display Priority - HIGH / MEDIUM / LOW
Case Type - REAL / RESEARCH / HYBRID
Full Dataset
The full dataset contains 5,000+ cases with additional fields:
- Scenario, Impact, Risk
- Root Cause, Recommendation
- Transferable Lesson, Key Pattern
- AI Confidence Score, Embeddings
🔗 Get the full dataset: aifi-project.vercel.app
Citation
AI Failure Intelligence Platform (2026).
AI Failure Intelligence Dataset.
https://huggingface.co/datasets/aifi-intelligence/ai-failure-intelligence
License
CC-BY-4.0 — Free to use with attribution.