Stats

Counts over the 34 records of dataset v0.3.0, taken verbatim from the upstream summary. Every figure is a number of records; a gap is a missing value upstream, never zero.

Verification status

Confirmed: 22
Reported: 10
Test / evaluation: 2

Severity

High: 18
Medium: 9
Critical: 5
Low: 2

AI role

Load-bearing: 18
Significant: 13
Incidental: 2
Disputed: 1

Incident category

AI-orchestrated campaign: 9
Infrastructure abuse / supply chain: 9
Autonomous attack: 8
Agent hijack / prompt injection: 6
Lab escape / evaluation: 2

Actor type

Unknown: 10
Cybercriminal: 6
Nation-state: 6
Researcher: 6
Lab test / evaluation: 3
Single operator: 3

Agentic autonomy level

Not applicable: 10
Supervised-autonomous: 8
Fully-autonomous: 7
Tool-assisted: 5
Unknown: 3
Human-in-the-loop: 1

Model family

Other / unspecified: 16
Claude (Anthropic): 14
Gemini (Google): 5
GPT (OpenAI): 5
Qwen (Alibaba): 1

Year of disclosure

2024: 2
2025: 14
2026: 18

Map coverage

Records with a stated location: 7 of 34
Points: 12 (12 country-level centroids)
By role: origin 9, target 3
By basis: Actor location 1, Sponsor attribution 8, Victim location 3

Source archiving

Source URLs: 69
With an archived copy: 61 (88%)