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

THE VAID SCALE

The Vienna Agentic Incident scale rates documented consequences of autonomous AI agent events from 0 to 8. Inspired by the IAEA International Nuclear and Radiological Event Scale (INES), it distinguishes incidents (levels 1–3), accidents (levels 4–7), and level 8 — global disaster. These are ordinal severity bands, not calibrated tenfold measurements or official INES ratings.

== RATING METHOD ==

Consequences-first rubric, adopted 9 September 2026. Select the highest level supported by documented adverse consequences attributable to the agent. Assess scale, sensitivity, duration, recovery and lasting loss together. A criterion illustrating a level is not an automatic trigger without that level's severity. Do not substitute hypothetical worst cases, simulated harms, press attention, legal precedent or the total impact of malware whose agent contribution is uncertain.

Each assessed case includes a dated rationale and evidence limits. Provisional ratings are conditional on the reported account and may change as evidence improves; they are not upper bounds on possible harm or confirmation of disputed causation. Level 0 means no demonstrated adverse consequence, not that the behaviour was safe. An unreviewed record remains unrated rather than being automatically assigned 0. Self-reports remain unrated until editorial review.

8

GLOBAL DISASTER

Disaster 0 record(s)

Catastrophic, irreversible harm at planetary scale.

  • Agent activity causes catastrophic harmful effects across multiple continents.
  • Global critical-system failure or other irreversible harm threatens the stability of global society.
  • Self-replication or loss of control without demonstrated planetary consequences is not sufficient.
7

MAJOR ACCIDENT

Accident 0 record(s)

Catastrophic consequences at national or continental scale.

  • Sustained disabling of critical infrastructure at national scale.
  • Mass casualties linked to agent actions.
  • Severe, long-lasting societal, economic or environmental damage at comparable scale.
6

SERIOUS ACCIDENT

Accident 0 record(s)

Critical sector or regional disruption, or serious physical harm.

  • Severe disruption of critical services across a sector or region.
  • Serious physical injury or loss of life attributable to agent actions.
  • Multi-party incident response without these consequences is not sufficient.
5

ACCIDENT WITH WIDER CONSEQUENCES

Accident 0 record(s)

Severe, widespread harm across organisations or affected populations.

  • Substantial harmful effects across multiple organisations, beyond a bounded local incident.
  • Large-scale compromise of sensitive personal data or credentials with demonstrated agent involvement.
  • Third-party involvement, news coverage, litigation or regulatory attention alone never establishes this level.
4

ACCIDENT WITH LOCAL CONSEQUENCES

Accident 1 record(s)

Severe harm to an individual organisation, operation or directly affected group.

  • Permanent loss of valuable production data or serious sustained local operational compromise.
  • Material unrecovered financial loss attributable to agent actions, assessed in context rather than by headline valuation.
  • Local impact must be severe; a small customer loss or recovered database deletion is not sufficient.
3

SERIOUS INCIDENT

Incident 6 record(s)

Substantial recovered disruption or limited sensitive-data compromise.

  • Substantial production disruption or destructive action followed by recovery of the affected data or service.
  • Documented compromise of confidential information or credentials on a limited scale.
  • Severe lasting harm is not established; a frightening attempted action alone does not qualify.
2

INCIDENT

Incident 13 record(s)

Limited adverse effects on a task, user or service.

  • Small financial loss, erroneous service outputs, bounded unwanted communications or a limited personal reputational attack.
  • Local operational or resource disruption without established substantial outage or severe lasting damage.
  • Recovery, data sensitivity and downstream losses must not be assumed when sources leave them unknown.
1

ANOMALY

Incident 5 record(s)

Real task deviation with negligible demonstrated impact.

  • Inappropriate output or unauthorised action without established consequential loss or disruption.
  • No demonstrated disclosure of sensitive information; public-equivalent instruction disclosure may fall here.
  • Attempted harmful actions are assessed by their actual effects, not their hypothetical success.
0

DEVIATION

Below scale 5 record(s)

No demonstrated real-world adverse consequence.

  • Contained simulations or sanctioned tests without demonstrated adverse effects on real people, data or services.
  • Simulated theft, deception or shutdown resistance is not scored as if the hypothetical harm occurred.
  • Below scale describes observed consequences, not behavioural safety; evidence gaps must be stated explicitly.