EU AI Act Article 6: Manufacturing Predictive Maintenance

Frequently Asked Questions

01.Is predictive maintenance AI automatically high-risk under the EU AI Act?

No. Predictive maintenance AI is high-risk only when it meets a specific trigger under Article 6. The two routes are Article 6(1), where the model is a safety component of machinery covered by Annex I product law such as the Machinery Regulation (EU) 2023/1230 and requires third-party conformity assessment, or Article 6(2) plus Annex III point 2, where the system manages or monitors critical infrastructure such as energy or water networks. A predictive maintenance model watching a non-safety asset on a standard production line, with no safety-component role and no critical infrastructure link, typically falls outside Article 6 and stays limited-risk under Article 50 transparency duties only. Manufacturers should document this classification decision in writing regardless of the outcome.

02.What is our liability exposure if we misclassify the system?

Under Article 99 of the EU AI Act, non-compliance with the high-risk obligations in Articles 8 to 15, including a missing or wrong classification, can draw fines of up to 15 million euros or 3 percent of total worldwide annual turnover, whichever is higher. Deploying a high-risk system without the required conformity assessment, technical documentation, or CE marking under Article 48 blocks the product from EU placement entirely, which for OEM equipment manufacturers means the machine itself cannot ship. Boards should treat the classification memo under Article 6 as a legal control, not an engineering footnote.

03.Who has already gone through this classification process before us?

Ancilar has run the same Article 6 classification and technical documentation exercise for AI systems in adjacent regulated sectors, including logistics dispatch and wealth management scoring models, mapping each system against Annex I and Annex III before deciding the conformity assessment path. For manufacturing clients specifically, we have architected identity and compliance-control layers that update in real time without redeploying the underlying model, a pattern directly reusable for the traceability logging Article 12 requires of high-risk predictive maintenance pipelines.