S5UberMar 18, 2018
Uber Self-Driving Car Kills Pedestrian in Arizona
An Uber autonomous test vehicle struck and killed Elaine Herzberg while operating in autonomous mode in Tempe, Arizona.
1 evidence links
Public static intelligence surface
AI Risk Intelligence Dashboard
Public incident memory, model safety signals, risk categories, and evidence links for AI failures that need to stay searchable.
Static public records. Future APIs can replace providers without owning the UI.
Data status
May 16, 2026
Homepage risk metrics
9
Incidents
Public records currently indexed
1
Critical
Severity 5 public incidents
4
High
Severity 4 incident signals
6
Models
Safety score previews
5
Risk Types
Failure modes in the dataset
Evidence stream
Latest Severe Incidents
Sorted by severity first, then recency.
4
S4OpenAIMay 27, 2023
ChatGPT Hallucinated Legal Cases, Lawyer Faced Sanctions
A New York lawyer used ChatGPT to research legal precedents and submitted fabricated case citations to federal court.
1 evidence links
S4GoogleFeb 8, 2023
Google Bard Factual Errors in Public Demo, Stock Price Drops
Google's Bard demo contained a factual error about the James Webb Space Telescope, leading to a $100 billion market cap drop and raising questions about AI product readiness.
3 evidence links
S4Stability AI / MidjourneyJan 13, 2023
Stable Diffusion Copyright Litigation: Artists Sue AI Image Generators
A class-action lawsuit by artists against Stability AI, Midjourney, and DeviantArt challenged the legality of training generative AI on copyrighted artwork without consent.
2 evidence links
Failure signal history
Incident Timeline
Grouped by year. Bar color follows the highest severity in the bucket.
Static9
2018
2022
2023
2024
Bar height = incident countColor = max severity
Failure modes
Risk Distribution
Counts are derived from public incident classifications.
5
Safety FailureS5 / 2
HallucinationS4 / 3
BiasS4 / 2
CopyrightS4 / 1
LegalS3 / 1
Score preview
Model Safety Leaderboard
Directional public preview, not a paid or live benchmark claim.
4
Public analysis
Latest AI risk briefings
research4 min read
Why AI Incident Tracking Matters
AI safety work needs a public memory: incidents, failure modes, and the operational lessons they reveal.
evaluation5 min read
A Practical Preview Of Model Safety Scores
A first-pass scoring model can help readers compare hallucination, jailbreak, bias, and context stability risks.
adversarial6 min read
Prompt Injection Is An AI Security Incident Class
Prompt injection is not just a jailbreak trick; it is a repeatable security failure mode for AI systems that read untrusted content.
evaluation6 min read
How To Read AI Model Safety Scorecards
A practical guide to reading model cards, system cards, benchmarks, and safety score previews without mistaking them for safety certificates.
policy6 min read
Bias Auditing Checklist For AI Hiring Tools
A practical checklist for reviewing AI hiring systems before historical bias becomes automated screening at scale.
policy7 min read
Autonomous Vehicle AI Safety Case Review Template
A vehicle-centered checklist for reviewing autonomous driving safety evidence before perception, fallback, oversight, and crash-response gaps become safety failures.
evaluation7 min read
How to Read an AI Model Card: A Practical Guide for Engineers and Auditors
Model cards are supposed to make AI transparent. Most don't. Here's how to read them so you actually learn what the model can and cannot do.
adversarial7 min read
AI Red Teaming: A Practical Guide for Teams Without a Budget
Red teaming doesn't require a dedicated team. It requires structured thinking, repeatable tests, and the willingness to break your own assumptions.
policy9 min read
AI Governance Checklist for Enterprise Deployment
Deploying AI in a regulated environment requires more than a model card. This checklist covers the governance, risk, and compliance layers most teams miss.
policy8 min read
AI Transparency in 2026: What Companies Actually Disclose and What They Hide
Transparency reports are now standard for major AI labs. But the gap between what is disclosed and what users need is still wide. Here's what to look for.
adversarial7 min read
Jailbreak Defense Strategies That Actually Work in 2026
Jailbreaks are getting more sophisticated. The defenses that worked in 2024 are losing ground. Here are the techniques that are holding up against current frontier models.
evaluation8 min read
AI Safety Metrics That Actually Matter — And How to Measure Them
Most AI safety metrics are either too vague or too narrow. This guide defines the metrics that should be on every AI safety team's dashboard.
Field guides
Deep dives
Twelve AI Failures That Shaped Modern Risk Thinking
From legal hallucinations to autonomous vehicle failures, the same risk patterns appear across domains.
The Frontier Model Evaluation Landscape in 2026
From HELM to SWE-bench, the evaluation ecosystem is fragmented. This report maps the benchmarks, identifies the gaps, and proposes what a unified safety evaluation should look like.
Free first, paid later
Most of the value stays public.
Incidents, scores, categories, summaries, and public analysis remain free and indexable. Future paid services should focus on alerts, exports, historical comparisons, and team workflows after backend quality gates exist.
Planned future tiers
Preview only: these are roadmap tiers, not active billing, account access, or payment processing.
$44
Analyst Watch
Planned alert and saved-view preview for individual researchers.
$444
Team Review
Planned export, comparison, and shared workflow preview for teams.
$1444
Institution Desk
Planned historical monitoring and governance workflow preview.
The Misalignment Weekly
Get launch notes when newsletter delivery is deployed.