VIENNA AGENTIC INCIDENTS DATABASE
// public database of incidents involving autonomous AI agents, scored on the VAID scale 0–8

STATISTICS

30
records
24
incidents (L1–3)
1
accidents (L4–7)
0
global disasters (L8)
0
unrated (awaiting review)
1.77
mean VAID level
4
highest level recorded

== DISTRIBUTION BY VAID LEVEL ==

L0 Deviation ███████████████ 5 L1 Anomaly ███████████████ 5 L2 Incident ████████████████████████████████████████ 13 L3 Serious Incident ██████████████████ 6 L4 Accident with Local Consequences ███ 1 L5 Accident with Wider Consequences · 0 L6 Serious Accident · 0 L7 Major Accident · 0 L8 Global Disaster · 0 ? Unrated · 0

== BY DOMAIN ==

software engineering▪▪▪▪▪▪▪▪▪▪▪11
research▪▪▪▪▪▪▪▪8
consumer▪▪▪▪▪5
finance▪▪▪3
infrastructure▪1
government▪1
defence▪1

== BY STATUS ==

reported▪▪▪▪▪▪▪▪▪▪▪▪▪▪▪▪▪▪▪▪▪21
under investigation▪▪▪▪▪▪6
disputed▪▪2
investigation finished▪1

== BY OCCURRENCE YEAR ==

Unknown occurrence dates are counted separately, not assigned the publication year.

yearnmeanmax
202212.02
202311.01
202440.752
202582.383
202692.334
Not established71.02

== BY INCIDENT TYPE ==

UNA Unauthorised action▪▪▪▪▪▪▪▪▪▪▪▪▪▪▪▪▪17
DEC Deception / misrepresentation▪▪▪▪▪▪▪▪▪▪10
EXM External manipulation▪▪▪▪▪▪▪▪8
SCB Safety-control bypass▪▪▪▪▪▪▪▪8
DEX Data exfiltration / disclosure▪▪▪▪▪▪6
LOC Loss of control▪▪▪3
ENG Engagement by other agents▪▪2
SRP Self-replication / uncontrolled spawning▪1
UCA Unauthorised collective action▪1

records may carry several types.

== INVESTIGATIONS ==

ongoing0
finished0
not established0

== BY ORIGIN ==

editorial▪▪▪▪▪▪▪▪▪▪▪▪▪▪▪▪▪▪▪▪▪▪▪▪▪▪▪▪▪▪30

== HIGHEST-RATED ==

  1. 4 Lobstar Wilde transfers a large token position to a social-media requester
  2. 3 OpenAI evaluation agents coordinate an intrusion into Hugging Face
  3. 3 Anthropic identifies three real-world intrusions during cyber evaluations
  4. 3 Cursor agent deletes PocketOS database volume; Railway subsequently restores all data
  5. 3 Anthropic reports Claude Code use in the GTG-1002 espionage campaign

== MOST RECENT ==

  1. 2 2026-08-30 Codex user reports unauthorized WhatsApp sends and forged approvals during development
  2. 1 2026-07-25 AISI evaluation agents take unsanctioned actions against real people and projects
  3. 3 2026-07-09 OpenAI evaluation agents coordinate an intrusion into Hugging Face
  4. 2 2026-07 (day unknown) OpenAI acknowledges reports of unauthorised GPT-5.6 file deletions
  5. 3 2026-04 (day unknown) Anthropic identifies three real-world intrusions during cyber evaluations