Manufacturing

Switzerland has no AI law of its own, and the Federal Council chose sector-by-sector rules on 12 February 2025.¹ If you build machines and sell them into the EU, EU rules reach you anyway, and sooner than most budgets assume.

The work that holds up on a production line is narrow and measured.

• A technician or line operator asks a question against existing manuals, maintenance history and incident logs. Our grounded knowledge copilot shows the pattern: it cites its sources or abstains. It is a working prototype.
• Where a basement or remote site has no reliable network, the model has to run on the device. See on-device inference, a working prototype.
• Our electric-boat helm was a client delivery that reached a first working prototype, wired into the vessel’s own systems over CAN bus.

Start with one equipment family and evaluate against technician questions before it reaches the floor. That gives you a baseline for the next family.

Under the AI Act, transparency obligations and enforcement structures apply from 2 August 2026. High-risk AI embedded in regulated products (Annex I) applies from 2 August 2028, after the June 2026 Digital Omnibus moved it back by a year.²

The Machinery Regulation applies from 20 January 2027, more than eighteen months before the Annex I date. It adds two AI categories to its own Annex I: machinery with safety components that use self-evolving machine learning, and machinery whose embedded AI ensures a safety function.³ An AI system that is a safety component under the Machinery Regulation is automatically a high-risk AI system under AI Act Annex I. A manufacturer of AI-enabled safety equipment therefore works to two clocks, and a budget built on the 2028 date alone misses the 2027 one.

Whether a system is built as a safety component at all decides which regime applies, so make that call during design.

This page is a screening note, not legal advice. Counsel should confirm how the two timelines combine for your product.

A pilot built on one machine’s manuals works. It is then extended to the whole fleet in one step, retrieval quality drops, and nobody notices until a technician gets a confident wrong answer. Keeping Stage 04 · Pilot in one place and Stage 05 · Scale what works separate catches this.

Start with the AI opportunity assessment. In one to two weeks, its regulatory applicability note answers the safety-component question before design begins. A proof of value sprint on one equipment family can follow. See also AI regulation in Switzerland.

• WorkA grounded knowledge copilot for field service
• WorkOn-device inference, with nothing leaving the device
• WorkA helm for a fully electric boat