Indices

Three live measurements of the AI agent security ecosystem. Each publishes alongside a methodology document explaining how the value is computed and how a third party can reproduce it.

Exposure Prevalence

Live · ARIAscout (Shodan / Censys)

Population-level measurement of exposed AI agent infrastructure on the public internet.

Preview. Values land Phase 2 (May 2026 target).

Attack Prevalence

Live · ARIAtrap (TrapMyAgent fleet, AgentPwn, HoneyFinder)

Volume and classification of observed attack sessions against the OpenA2A honeypot fleet, partitioned by Threat Matrix technique.

Preview. First measurement window after one full month of fleet telemetry.

Sophistication

Live (distribution-only) · ARIAtrap session content

Technical sophistication of observed attacks on a one-to-ten scale. Distribution-only until classifier validation.

Preview. Distribution only. Mean withheld pending classifier reaching F1 threshold on labeled holdout.

Why three indices, not one

Exposure, attack volume, and sophistication measure different things. Collapsing them into a single ecosystem score would obscure the question being asked. Exposure is observable from outside. Attack volume is observable on instrumented surfaces. Sophistication is observable only when sessions are deep enough to classify. Each index publishes independently with its own methodology and its own confidence interval.