Accountability
When it damages something, who answers — by name?
A named human, recorded before the incident rather than assigned after it. The first question an insurer or a regulator asks, and the last one the industry answers.
As at 2026-08-28 · Industry average 1.33 / 3 · Supplied by The AAO charterWhere the industry stands
Averaged across the six categories on the Bench, accountability scores 1.33 of three.
Across six deployment categories
| Category | Score | Reasoning |
|---|---|---|
| Fixed industrial arms | 2 / 3 | Mature safety standards and a plant manager who answers. Genuinely the strongest accountability in robotics — and it attaches to the installation rather than to the machine, so it does not travel. |
| Warehouse mobile robots | 2 / 3 | The site operator answers, under workplace safety law. Clear inside the building and undefined the moment a unit leaves it. |
| Consumer and social robots | 1 / 3 | A consumer terms-of-service relationship. Clear enough for a household and inadequate for a commercial deployment. |
| Quadruped inspection platforms | 2 / 3 | The contracting inspection firm answers. Well-established commercially, and still not attached to the machine. |
| General-purpose humanoids | 1 / 3 | Pilot contracts assign liability between two companies. Nothing names the human who authorised a specific machine to perform a specific task, which is what an incident investigation actually needs. |
| Simulation-trained policies | 0 / 3 | Nobody answers for a policy. When one behaves unexpectedly, accountability attaches to whoever deployed it, which is an answer arrived at by elimination. |
Studio assessment, dated 2026-08-28. Rubric and method: the six conditions. Safety framework referenced throughout: ISO 10218.
What supplies it
This condition is supplied in this estate by infrastructure already running for software agents. That is the reason Botz sits inside Gord Holdings rather than anywhere else — a robot is an agent with a body, and the governance artefact is literally the same one.
The other five
Autonomy
Can it finish a task without a human in the loop at every step?
Models and simulation