Detection Model Training

A detector earns the right to alert on your data.

AI Detection · Updated Sep 24, 2026

How does a detector earn an alert?

Cyberdroid AI Detection runs 21 ML detectors. Each one is proven on your own data, in shadow, before it can raise a single alert.

See the method

Method

Every detector is disabled, shadow or production. Only production detectors raise alerts.

Shadow findings go to a separate channel, so a new detector is measured without reaching your queue.

Replay tooling measures precision, recall, stability and volume before promotion.

Promotion and rollback are recorded, so every change in what alerts has a history.

Emergency caps per detector and per tenant route overflow to shadow instead of your queue.

Measures

Precision, recall, stability and volume on replay of your telemetry. These decide promotion. We publish the method, not a score.

Limits

Staging cannot anticipate behavior your telemetry has not shown yet. A detector proven on replay is measured, not guaranteed.

Sources

Related

Unsupervised Behavioral Modeling

Method

Evaluations and Benchmarks

Method

AI Detection

Product
Contact
ARRTECH

© 2026 ARRTECH Corporation. All rights reserved.