HUMAN-CONTROLLED AI

AI-use notice

Version ai-use-2026-08-14 · Controlled preproduction pilot

Approval status: preproduction draft. The configured use cases, data classes, providers, regions, retention, evaluations, and human-review controls require formal approval before production.

What AI may do

ARIA may help identify candidate equipment or components, assess capture quality, transcribe authorized narration, and draft evidence-linked observations or proposed dispositions. The configured feature may send the minimum authorized inspection context to an approved AI provider.

What AI may not do

AI output is not an authoritative measurement, qualified NDT result, engineering approval, customer commitment, or released quote. It cannot silently change equipment/component identity, approve its own output, or bypass an authorized technical reviewer.

Your responsibilities

Confirm equipment and component identity, inspect source evidence, correct errors, record limitations, protect unrelated personal or commercial speech, and use manual or qualified methods where required. If evidence is missing, contradictory, low quality, or outside validated coverage, request review or recapture.

Data-use boundary

This acceptance permits the configured AI features to process authorized evidence for the current tenant workflow. It does not permit model training, shared/global learning, cross-customer reuse, marketing, or sale. Those uses require a separate organization-level data-use approval and controlled deployment.

Audit and withdrawal

ARIA records the policy version and acceptance. Your administrator can suspend access or AI features. Withdrawing acceptance blocks inspection-data access until policy and workflow requirements are resolved; it does not automatically delete records subject to quality, security, retention, or legal-hold controls.