Skip to content

Responsible deployment

How Cyberdroid AI detectors move from testing into production.

How models reach production

Evaluate models in stages on representative activity, and retain the reason for each production change. The team should remain able to inspect performance and withdraw a detector when needed.

A detector can be disabled, evaluated in shadow or promoted to production. Promotion and rollback decisions are recorded; output limits keep overflow available in shadow.

Stages

  1. Three detector states

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

  2. Replay testing

    Before promotion, replay shows how accurate, complete, steady and noisy a detector is.

  3. Written criteria

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

Limits

Unseen activity

Evaluation reflects the data and cases tested. Continue reviewing source quality and new behavior, with rollback available when results warrant it.

Status

Cyberdroid AI Detection is available now to select organizations. Cyberdroid AI Investigation is in early access.

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.

ISO

ISO/IEC 42001 Annex A area A.6 covers the AI system life cycle.

ARRTECH

© 2026 ARRTECH Corporation. All rights reserved.