AI systems & AI Act scope

If your machinery embeds AI, the question your customers (and eventually authorities) will ask is: is it a high-risk AI system? For machine builders the honest answer is usually no — and Resilic is built to say so, with the reasoning attached.

The inventory

Per product, record each embedded AI system: name, purpose, supplier, notes — and two questions. Both are deliberately tri-state (yes / no / not answered); an unanswered question is never defaulted, because the verdict depends on it:

1. Does it ensure a safety function?
2. Does it have fully or partially self-evolving (machine-learning) behaviour?

These two questions mirror the narrow trigger for the machinery route: under AI Act Art. 6(1), an AI system is high-risk via this route when it is a safety component (or the product itself) under listed Union harmonisation legislation — the Machinery Regulation is listed — and subject to third-party conformity assessment. MR Annex I Part A items 5–6 (safety components / machinery with self-evolving behaviour) are what force that notified-body route for machinery.

The verdicts

The classification is derived on read — never stored — so a logic or wording fix never leaves a stale verdict. Each result shows a numbered, cited decision path:

• Out of scope (machinery route) — not a safety function, so the Art. 6(1) trigger is not met. The result notes that AI Act Annex III (standalone high-risk use contexts) is a separate check outside this classifier.
• Out of scope — verify MR route — a safety function without self-evolving behaviour sits outside the items-5–6 mandatory notified-body category, but the applicable MR conformity route still has to be verified independently. The verdict says so rather than pretending certainty.
• High-risk (machinery route) — safety function and self-evolving behaviour: third-party conformity assessment under MR Annex I Part A items 5–6, hence high-risk under Art. 6(1). Provider duties apply from 2 Aug 2028 for product-embedded high-risk AI.
• Incomplete — a question is unanswered; the verdict names exactly which one.

Honesty framing

Most machine-builder AI — predictive maintenance, non-safety vision QC, process optimization — is not high-risk, and “you are probably out of scope, and here is why” is the answer Resilic gives when the facts support it. Every verdict carries the same boundary statement: this is a readiness-scope classification based on your answers. Resilic is not a notified body, does not certify, and a verdict here is not a conformity determination.