Inside the Room: What Happened When Europe's First Humanoid Safety Roundtable Met in Zurich
In early June 2026, HRAS sat in a room in Schlieren with researchers, engineers, and risk professionals, all there to answer one question: what does it actually take to deploy a humanoid robot safely?
Nobody in the robotics world wants to answer that question properly. It's easier to talk about capability. Speed. The next demo video.
We kept the group small. Deliberately so. Voices from ETH Zurich, EMPA, industry, logistics, and the insurance sector — people who speak from hands-on experience, not press releases.
One thing, zero debate
Here's the thing about safety roundtables: they usually produce vague agreement and sharp disagreement. This one flipped that.
On one point, nobody pushed back. Every deployed humanoid needs a physical, accessible emergency stop. Not a setting buried three menus deep in some app. A red button. Something anyone in the room can reach, including someone who's never touched the robot before. Sounds obvious, right? It isn't standard yet. And honestly, that's the real story here. The easiest point of agreement in the room wasn't about some exotic edge case. It was about the basics. That tells you exactly where the industry's gaps are.
Unfinished Machines
Midway through, someone offered a phrase that reframed the entire conversation: Unfinished Machines.
That's what today's humanoids are. Not finished products with fully mapped behaviour. Systems still discovering their own limits — what they can do, where they fail, how they behave under pressure they haven't seen before.
We think this is the most useful idea to come out of the whole session.
Why? Because it changes the question. You don't govern an unfinished system the way you'd govern a finished one. You bound it. Not because the technology is bad — because deploying something that's still revealing its own failure modes, without limits, is exactly how avoidable incidents happen.
A "licence to operate," borrowed from aviation
So if a humanoid is unfinished, you don't just unbox it and flip the switch. We need something closer to how aviation handles airworthiness. The group proposed a licence to operate. A structured evidence package, reviewed before deployment, tied to one specific system, one environment, and one defined task. Change any of those — swap the robot model, move it to a different facility, or alter the job it's doing — and the licence gets reassessed. No exceptions.
What's in that evidence package? Some of it is what you'd expect — risk assessments, robustness testing. But the part people skip is more interesting: a clear, honest statement of what the system cannot do. Documented fail-safe steps. Verified proof that the operators actually know what they're doing.
Who's responsible when something goes wrong?
This is the question that's been hanging over the entire industry. Builder, installer, or operator — who's actually on the hook?
We landed on a clean three-tier answer:
- OEM — owns the hardware, the base software, the security architecture, the documented limits.
- System integrator — owns safe deployment: site validation, operator training, making sure the environment actually fits the machine.
- Operator — owns correct use within those bounds. Override a safety control manually, and the accountability shifts to you.
Why a neutral body has to exist
No manufacturer can credibly certify its own product. Let's be honest about that. And no operator can credibly audit themselves either.
Every safety ecosystem that actually works — aviation, automotive, industrial machinery — eventually needed an independent party with zero stake in the commercial outcome. Think TÜV. That role doesn't exist yet for humanoid robots.
The roundtable handed that mandate to HRAS. Not as a regulator. Not as a vendor. As a neutral inspection and certification authority for humanoid deployment in Europe. That's a big responsibility, and we're taking it seriously.
This was session one
Everything above is the headline version. The real working document goes much deeper — a complete eight-domain governance baseline, the full evidence framework, the specific events that trigger a mandatory audit, and twelve action items we're already executing, each with an owner and a deadline attached.
The full Working Session Outcome Report is for HRAS members and subscribers. If humanoid deployment touches your work in any way — and increasingly, it touches more work than people expect — this is the moment to be in the room.
Get the full report
HRAS members and subscribers get the complete Working Session Outcome Report, plus ongoing access to working group findings as we build out the governance framework over the coming months.
HRAS (Humanoid Robotics Association Switzerland) is a Zurich-based association focused on governance, certification, and deployment readiness for humanoid robots in Swiss and European environments.