The Automotive Industry Action Group has published CQI-38, a new guideline for evaluating AI-based vision inspection systems on automotive production lines. The document gives OEMs, suppliers at every tier, and vision inspection equipment manufacturers a standardized way to assess how capable, accurate, and reliable an AI vision system is before it goes live on the shop floor.
CQI-38 is built to supplement IATF 16949 rather than replace it. It applies specifically to systems that perform feature recognition and defect classification from image data. It does not cover traditional rule-based inspection or systems doing precise dimensional measurement. That distinction matters because AI vision models behave differently from fixed-logic inspection tools: their outputs can shift as training data, lighting, or camera setups change, and CQI-38 treats that variability as a risk to manage rather than an edge case.
What's new here is the lifecycle framing. Instead of a one-time validation checklist, CQI-38 covers system planning and manufacturing implementation, system and process acceptance, capability maintenance, and continual improvement. It also sets explicit requirements for controlling changes to equipment, AI models, training data, and the operating environment.
For a quality engineer or CMM operator, that maps onto a familiar problem: an inspection system that passed validation months ago but has since seen a lighting change, a new lens, or a retrained model. CQI-38 gives auditors and internal quality teams a shared vocabulary and set of checkpoints for deciding whether that kind of drift requires revalidation, rather than leaving it to individual plant judgment.
AIAG is distributing CQI-38 as an e-document in multiple languages, with a hardcopy version bundled with a downloadable assessment tool, and a separate online or offline viewer with downloadable files. Buyers outside the U.S. looking for print copies are directed to AIAG's network of authorized distributors rather than a direct hardcopy purchase.
