AI Incident Tracker
This work is also hosted by the MIT AI Risk Repository.
All incidents in the AI Incident Database have been processed using an LLM and classified according to the MIT Risk Repository causal and domain taxonomies. The severity of harm and National Security Impact have been assessed for each incident.
This is intended as a proof of concept to explore the potential capabilities and limitations of a scalable incident analysis framework.
This blog post discusses the background, the approach taken, preliminary results and next steps.
Please feel free to explore the analysis through the dashboard pages below and share feedback.
What’s New? Major Update - 23 June 2025
Latest incidents from AIID added
This brings us completely up to date with the latest incidents on the AI Incident Database as at 23 June 2025 (up to incident ID #1116)
LLM Analysis
Some updates have been made to how the tool processes reports in order to improve validity of analysis.
The full dataset has been classified using this latest iteration of the tool (reclassification of all incidents that were classified with the previous version, to ensure consistency across the dataset).
Harm Severity
The harm severity analysis uses a new scale, simplifying scoring to 1-5 points (previously 0-10).
I have tried to remove ambiguities from the definitions in the scale and to make it possible to grade incidents consistently with greater objectivity.
Impact Profile - visualises the reported harm caused in each category as a spider-chart, making it easy to compare incidents or identify certain profiles of interest.
National Security Impact Assessment
Assesses NatSec impact of each incident in 5 categories: Physical Security & Critical Infrastructure / Information Warfare & Intelligence Security / Sovereignty & Government Functions / Economic & Technological Security / Societal Stability & Human Rights using this framework.
Classifies threat for Imminence, Novelty and Autonomy
NatSec Incident View presents the NatSec Impact Profile as a spider chart
Potential Causes of Incident
A Fishbone/Ishikawa Diagram presents a number of potential causes for each incident, organised by category. (Incident View)
Ambiguities and Alternative Interpretations
‘Ambiguities identified’ and ‘Alternative interpretations’ - makes it easier to spot analyses where further review of the reports or investigation is required
Primary Goal of AI Systems
Now includes the primary goal of the AI system involved in each incident, classified according to a taxonomy based on the AIID GMF. (Risk Classification)
Impact Profile
Explore the AI Incident Tracker:
Risk Classification - distribution of incident classifications (causal and domain) across the entire dataset
Incident View - classification and harm severity scores for each individual record, including summary of reasoning and confidence in analysis. This supports filtering by any combination of data fields.
Timeline: Risk Classification - distribution of incident classifications by year
Timeline: Sub-domains - distribution of incident sub-domains by year
Timeline: High Severity Incidents - incidents with high direct harm severity scores by year
Timeline: High Severity Multiple Categories - incidents causing severe harm in more than one harm category
Timeline: Direct Harm Caused - distribution of harm severity scores by year
NatSec Incident View - national security impact assessment for each individual incident, including summary of reasoning
Timeline: NatSec Impact - distribution of national security impact of incidents by year
Harm Severity Scale - scale to assess severity of harm cased across 10 categories
NatSec Impact Assessment Framework - scale to assess impact over 5 categories and identify threat characteristics (novelty, autonomy, imminence).
