Responsible Deployment

No machine-learning detector alerts until it is measured on your data.

AI Detection and AI Investigation · Updated Sep 23, 2026

How does a model reach production?

In stages, on your data first. Each Cyberdroid AI Detection detector is measured in shadow before a recorded promotion lets it alert. Cyberdroid AI Investigation reaches customers only through early-access evaluations with written criteria.

Diagram: a detector moves from disabled to shadow, where it is measured, then by promotion to production, where it alerts. A dashed path returns from production to shadow on rollback or overflow.

Stages

  1. Three states.

    Each of the 21 detectors is disabled, shadow or production. Only production alerts.

  2. Measured on replay.

    Precision, recall, stability and volume are measured before promotion.

  3. Criteria in writing.

    Cyberdroid AI Investigation runs as an evaluation, with success and stop criteria agreed first.

Limits

Guarantee

Shadow shows how a detector behaves on your data so far, not on activity it has not yet seen. That is why rollback exists.

Status

AI Investigation is not generally available. Its only path to your environment is a design-partner evaluation.

Standards

Naming a clause is not a claim of certification.

NIST

AI RMF MANAGE 2.4 asks for a way to disengage a system, and MANAGE 4.1 for post-deployment monitoring. Recorded rollback and overflow to shadow answer both.

ISO

ISO/IEC 42001 Annex A area A.6 covers the AI system life cycle. The three detector states are that life cycle, run in your environment.

Related

Evaluations

Autonomy Levels

System Cards

Cyberdroid AI Detection

Contact
ARRTECH

© 2026 ARRTECH Corporation. All rights reserved.