The real risk is not thinking machines. It is agents without consciousness.
Why autonomous systems need a common language of risk, and why that language has to be open.
A new generation of systems is operationally autonomous and morally empty. They plan, pay, move goods and work next to people. They complete the task without understanding its consequences.
When such a system causes harm, responsibility dissolves between the developer of the model, the integrator, the operator and the data provider. Each of them can point to the others. The injured party waits for years, and the insurer cannot price what it cannot measure.
Society can no longer accept “the algorithm decided” as an answer. Every autonomous action needs a human or an organisation that stands behind it, and financial cover that is agreed before something goes wrong, not negotiated afterwards.
This is not a question of science fiction or machine consciousness. It is a question of plumbing: shared definitions, verifiable evidence and clear roles. Aviation, shipping and medicine built that plumbing over decades. Autonomous systems are spreading faster, so the work has to start now and it has to be done in the open.
RoboIns exists to do this work as a neutral, non-profit association, independent of any single vendor, insurer or platform.
- Accountability is human. Every autonomous system has a named responsible party.
- Risk is measured, not guessed. Objective parameters replace missing loss history.
- Evidence is verifiable. A trigger counts only if an independent source can confirm it.
- Basis risk is disclosed. Buyers know what a parametric policy does not cover.
- The standard is open. Its text is public and every version is consulted on.
- Risk stays with those licensed to carry it. RoboIns sets the standard; licensed insurers carry the risk.
- No single interest dominates. Industry, insurers, science and civil society share governance.
Four things public authorities can do now
Addressed to the European Commission, the Swiss Federal Council and FINMA, and to national supervisors working on AI and product liability.
Recognise a common incident format
Accept a standard Agent Incident Record as evidence under the EU AI Act logging duties and the revised Product Liability Directive (EU) 2024/2853.
Support a neutral registry
Encourage anonymised reporting of incidents involving AI agents and robots to an independent registry, as aviation does with safety reports.
Clarify parametric cover
Give licensed insurers legal clarity for parametric products based on open, verifiable triggers, including disclosure of basis risk.
Fund open work
Treat open risk standards for autonomous systems as public infrastructure eligible for research and innovation funding.
Do you share these principles?
We invite researchers, insurers, operators and public bodies to sign up as supporters and to help shape the first version of the standard.