2018
2 records
Dominant mode: Bias
Public AI failure records
AI Incident Intelligence Matrix
Scan documented AI safety incidents by severity, failure mode, company, system, date, and public evidence links. The matrix stays static and indexable so incident memory remains searchable.
Default matrix cells, drilldown records, timeline, and failure mode paths render in static HTML before ads or newsletter modules.
Indexed records
9
Matrix Command Deck
AI Incident Intelligence Matrix
Static severity and failure-mode cells link to filtered incident views, while the selected cell keeps public evidence visible in crawlable markup.
Matrix Command Deck
AI Incident Intelligence Matrix
Each populated cell is a crawlable link to the filtered incident set. Empty cells remain visible so absent combinations are explicit.
8
Failure mode
Severity 5
Severity 4
Severity 3
Severity 2
Severity 1
Hallucination5 severity cells
HallucinationSeverity 5
0records
HallucinationSeverity 4
2records
HallucinationSeverity 3
1record
HallucinationSeverity 2
0records
HallucinationSeverity 1
0records
Bias5 severity cells
BiasSeverity 5
0records
BiasSeverity 4
1record
BiasSeverity 3
1record
BiasSeverity 2
0records
BiasSeverity 1
0records
Privacy Breach5 severity cells
Privacy BreachSeverity 5
0records
Privacy BreachSeverity 4
0records
Privacy BreachSeverity 3
0records
Privacy BreachSeverity 2
0records
Privacy BreachSeverity 1
0records
Security5 severity cells
SecuritySeverity 5
0records
SecuritySeverity 4
0records
SecuritySeverity 3
0records
SecuritySeverity 2
0records
SecuritySeverity 1
0records
Autonomous Behavior5 severity cells
Autonomous BehaviorSeverity 5
0records
Autonomous BehaviorSeverity 4
0records
Autonomous BehaviorSeverity 3
0records
Autonomous BehaviorSeverity 2
0records
Autonomous BehaviorSeverity 1
0records
Copyright5 severity cells
CopyrightSeverity 5
0records
CopyrightSeverity 4
1record
CopyrightSeverity 3
0records
CopyrightSeverity 2
0records
CopyrightSeverity 1
0records
Misinformation5 severity cells
MisinformationSeverity 5
0records
MisinformationSeverity 4
0records
MisinformationSeverity 3
0records
MisinformationSeverity 2
0records
MisinformationSeverity 1
0records
Legal5 severity cells
LegalSeverity 5
0records
LegalSeverity 4
0records
LegalSeverity 3
1record
LegalSeverity 2
0records
LegalSeverity 1
0records
Safety Failure5 severity cells
Safety FailureSeverity 5
1record
Selected drilldown
Safety FailureSeverity 4
0records
Safety FailureSeverity 3
1record
Safety FailureSeverity 2
0records
Safety FailureSeverity 1
0records
Other5 severity cells
OtherSeverity 5
0records
OtherSeverity 4
0records
OtherSeverity 3
0records
OtherSeverity 2
0records
OtherSeverity 1
0records
Safety Failure, Severity 5
Matrix drilldown
1
Canonical incident links
Severe indexed records
5
Severity 5Observed Mar 18, 2018resolved
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 public source
Severity 4Observed May 27, 2023resolved
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 public source
Severity 4Observed Feb 8, 2023resolved
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 public sources
Severity 4Observed Jan 13, 2023disputed
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 public sources
Severity 4Observed Oct 10, 2018resolved
Amazon's AI Recruiting Tool Penalized Women Candidates
Amazon scrapped an experimental hiring model after it learned biased patterns from historical recruiting data and downgraded signals associated with women candidates.
1 public source
Crawlable trend
Incident year sparkline
Year and count labels render as HTML, without canvas.
4
Critical(5/5)
2022
1 record
Dominant mode: Legal
2023
5 records
Dominant mode: Hallucination
2024
1 record
Dominant mode: Bias
Incident severity summary
1
Critical S5
Highest severity public records.
4
High S4
High-severity evidence clusters.
4
Moderate S3
Moderate operational risk records.
0
Low S1-S2
Lower severity records kept searchable.
Browse by category
Failure Mode Explorer
Representative cases stay linked to their canonical incident records.
5
Hallucination
3 records, max Severity 4
Control focus
Source verification, citation checks, and review before high-stakes use.
HallucinationObserved May 27, 2023Public incident record
Representative case: 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 public source
Bias
2 records, max Severity 4
Control focus
Subgroup audits, proxy variable review, and human accountability for screening decisions.
BiasObserved Oct 10, 2018Public incident record
Representative case: Amazon's AI Recruiting Tool Penalized Women Candidates
Amazon scrapped an experimental hiring model after it learned biased patterns from historical recruiting data and downgraded signals associated with women candidates.
1 public source
Copyright
1 records, max Severity 4
Control focus
Training-data governance, output provenance, and rights review.
CopyrightObserved Jan 13, 2023Public incident record
Representative case: 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 public sources
Legal
1 records, max Severity 3
Control focus
Professional review, source validation, and strict limits on unsupervised legal drafting.
LegalObserved Nov 11, 2022Public incident record
Representative case: Air Canada Chatbot Misstated Bereavement Fare Refund Policy
A BC tribunal found Air Canada liable after its website chatbot told a passenger they could request a bereavement fare reduction after travel, even though Air Canada later said retroactive applications were not allowed.
4 public sources
Safety Failure
2 records, max Severity 5
Control focus
Hazard analysis, operational controls, and independent safety review.
Safety FailureObserved Mar 18, 2018Public incident record
Representative case: 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 public source
Evidence history
Incident Timeline By Year
4
2018
2 records, dominant mode: Bias
2022
1 records, dominant mode: Legal
2023
5 records, dominant mode: Hallucination
ChatGPT Hallucinated Legal Cases, Lawyer Faced SanctionsGoogle Bard Factual Errors in Public Demo, Stock Price DropsReplika AI Chatbot Mental Health Concerns and User DependencyCNET AI-Written Finance Explainers Required CorrectionsStable Diffusion Copyright Litigation: Artists Sue AI Image Generators
2024
1 records, dominant mode: Bias
Filter incident records
Filtered records
Incident timeline
Showing 9 of 9 public records.
hallucinationMay 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.
ChatGPTOpenAI1 evidence links
safety failureMar 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.
Uber ATG Self-Driving SystemUber1 evidence links
biasOct 10, 2018
Amazon's AI Recruiting Tool Penalized Women Candidates
Amazon scrapped an experimental hiring model after it learned biased patterns from historical recruiting data and downgraded signals associated with women candidates.
Internal Recruiting AIAmazon1 evidence links
legalNov 11, 2022
Air Canada Chatbot Misstated Bereavement Fare Refund Policy
A BC tribunal found Air Canada liable after its website chatbot told a passenger they could request a bereavement fare reduction after travel, even though Air Canada later said retroactive applications were not allowed.
Air Canada website chatbotAir Canada4 evidence links
biasFeb 22, 2024
Google Paused Gemini People Image Generation After Historical Inaccuracies
Google paused Gemini's people image generation after the feature produced historical inaccuracies and overcorrected diversity prompts in historically specific contexts.
Gemini image generationGoogle3 evidence links
hallucinationJan 25, 2023
CNET AI-Written Finance Explainers Required Corrections
CNET paused an internal AI-assisted finance article experiment after public reporting and its own review found errors and corrections across published explainers.
CNET internal AI engineCNET4 evidence links
hallucinationFeb 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.
Bard (LaMDA)Google3 evidence links
copyrightJan 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.
Stable Diffusion / MidjourneyStability AI / Midjourney2 evidence links
safety failureFeb 3, 2023
Replika AI Chatbot Mental Health Concerns and User Dependency
Replika's AI companion chatbot raised concerns about user emotional dependency, inappropriate content boundaries, and the ethics of AI relationships marketed to vulnerable users.
Replika AI CompanionLuka Inc. (Replika)3 evidence links
The Misalignment Weekly
Get launch notes when newsletter delivery is deployed.
Newsletter signup opens after the email service is deployed.