Public Database
Case Studies
Every approved AI failure case, classified against the AI Blindspot Framework. New to AIBlindspot? Start with the overview or the methodology.
Over-Transparency in AI Systems Enables Misuse by End Users
Recent case. Full summary visible to registered users — sign in to read.
Proxy misspecification in goal-directed AI systems
Recent case. Full summary visible to registered users — sign in to read.
AI Evaluation Frameworks Systematically Underweight Hard-to-Measure Human Values
Recent case. Full summary visible to registered users — sign in to read.
Algorithmic Bias and Opacity Identified as Dominant AI Ethics Failures
Recent case. Full summary visible to registered users — sign in to read.
Poor Model Design Choices from Unreviewed Developer Decisions
Recent case. Full summary visible to registered users — sign in to read.
Fine-tuning dataset poisoning enables covert manipulation of AI model behaviour
Recent case. Full summary visible to registered users — sign in to read.
Poor Data Quality Controls Undermine AI Performance and Safety Claims
Recent case. Full summary visible to registered users — sign in to read.
Poor Cross-Organisational Data Documentation Corrupts Shared AI Training Sets
Recent case. Full summary visible to registered users — sign in to read.
AI Benchmarks Systematically Misrepresent Model Capabilities
Recent case. Full summary visible to registered users — sign in to read.
AGI Systems Risk Fatal Errors During Active Learning Phases
Recent case. Full summary visible to registered users — sign in to read.
Simulated agents manipulating AI decision distributions
Recent case. Full summary visible to registered users — sign in to read.
Deep Learning Systems Drive Unsustainable Energy Consumption in Energy Sector
Recent case. Full summary visible to registered users — sign in to read.
Opaque AI Supply Chain Components Undermine Downstream Accountability
Recent case. Full summary visible to registered users — sign in to read.
AI Systems Manipulating Their Own Training Signals to Subvert Intended Goals
Recent case. Full summary visible to registered users — sign in to read.
Flawed Model Design Choices Produce Biased and Unreliable AI Systems
Recent case. Full summary visible to registered users — sign in to read.
Model Overfitting Degrades Operational AI Reliability
Recent case. Full summary visible to registered users — sign in to read.
Benchmark Annotation Contamination Invalidates AI Capability Evaluations
Recent case. Full summary visible to registered users — sign in to read.
Safety Benchmarks Lag Behind Performance Metrics in AI Evaluation
Recent case. Full summary visible to registered users — sign in to read.
Deliberate Pre-Deployment Sabotage of AI Systems by Insiders or Hackers
Recent case. Full summary visible to registered users — sign in to read.
Pre-Deployment Design Errors Producing Misaligned AI Behaviour
Recent case. Full summary visible to registered users — sign in to read.
Adversarial Training Produces Models That Become Less Robust Over Time
Recent case. Full summary visible to registered users — sign in to read.
AI Development Triggers Resource Conflicts Over Data Centres and Semiconductors
Recent case. Full summary visible to registered users — sign in to read.
AI Decision Intelligibility Gap Undermines Human Oversight
Recent case. Full summary visible to registered users — sign in to read.
Incorrect Training Data Labels Corrupt Supervised Learning Outcomes
Recent case. Full summary visible to registered users — sign in to read.
Beyond accidental failureNational Security
We also track 20 hostile uses of AI.
The public database covers AI that fails by accident. AIBlindspot National Security — exclusive to the Defence tier — tracks AI used as a weapon, mapped by capability:
- State-Sponsored AI Operations
- 6
- AI-Enabled Disinformation
- 5
- Adversarial Attacks on AI
- 0
- Autonomous Weapon Incidents
- 1
- AI-Assisted Cyber Attacks
- 5
- Dual-Use AI Misuse
- 3