These visualizations are experimental and still being refined. Designs and data presentation may change.

Causes by Subdomain (Heatmap)

Percentage of risks in each subdomain classified by causal taxonomy. Rows show domains and subdomains; columns are grouped by Entity, Intent, and Timing. Darker cells indicate higher concentration.

Key Takeaways

1.Across 1595 risks, the dominant causal entity is AI system (41%), intent is Unintentional (34%), and timing is Post-deployment (59%).
2.The most concentrated subdomain is 4.3 Fraud, scams, and targeted manipulation where 94% of risks are classified as "Post-deployment".
3.All 24 subdomains are shown across 3 causal dimensions (Entity, Intent, Timing), each with 3 coded values.
4.Hover over cells to see exact counts. Click any cell or subdomain name to filter risks.

Causal taxonomy classifications from the MIT AI Risk Repository.

	Entity			Intent			Timing
Subdomain	Human	AI system	Other	Intentional	Unintentional	Other	Pre-deployment	Post-deployment	Other
1 Discrimination & Toxicity	10%	61%	10%	5%	46%	31%	9%	57%	15%
2 Privacy & Security	56%	27%	15%	48%	30%	19%	18%	56%	24%
3 Misinformation	9%	72%	19%	8%	52%	40%	3%	76%	21%
4 Malicious Actors & Misuse	75%	10%	13%	84%	1%	13%	1%	89%	8%
5 Human-Computer Interaction	37%	24%	40%	12%	50%	38%	0%	79%	21%
6 Socioeconomic & Environmental	48%	25%	23%	23%	40%	33%	17%	45%	34%
7 AI System Safety, Failures & Limitations	12%	69%	17%	31%	40%	27%	15%	49%	34%

	Entity			Intent			Timing
Subdomain	Human	AI system	Other	Intentional	Unintentional	Other	Pre-deployment	Post-deployment	Other
1 Discrimination & Toxicity	10%	61%	10%	5%	46%	31%	9%	57%	15%
2 Privacy & Security	56%	27%	15%	48%	30%	19%	18%	56%	24%
3 Misinformation	9%	72%	19%	8%	52%	40%	3%	76%	21%
4 Malicious Actors & Misuse	75%	10%	13%	84%	1%	13%	1%	89%	8%
5 Human-Computer Interaction	37%	24%	40%	12%	50%	38%	0%	79%	21%
6 Socioeconomic & Environmental	48%	25%	23%	23%	40%	33%	17%	45%	34%
7 AI System Safety, Failures & Limitations	12%	69%	17%	31%	40%	27%	15%	49%	34%