When AI Acts, Who Answers? A Legal Framework for the Agentic Age
Who is liable when AI acts on its own?
A German court just held a clinic liable for false information its AI chatbot published, even though the system had been correctly configured. The operator could not hide behind the machine.
This is not an isolated ruling. It is the first signal of a deeper shift in how courts will treat autonomous AI systems, and existing legal frameworks are not ready for it.
The question is no longer what did the AI do. It is who gave it the keys : the data it could access, the transactions it could execute, the systems it could reach. Whoever grants those permissions owns what follows. Permission grants are documented, dated, attributable. The code is not.
In the article below, Crystal Dubois (Bonnard Lawson) and Eva Thelisson (AI Transparency Institute) map where Swiss civil liability frameworks fail to capture autonomous AI agents, in contract, in product liability, in tort, and propose three concrete steps organisations can adopt now, without waiting for new legislation.
The conference report synthesizes the findings of the Second International Conference on Safe and Ethical AI, convened by IASEAI at UNESCO House in Paris in February 2026.
The conference brought together leading researchers, engineers, regulators, and legal experts to confront a single urgent question: as AI systems evolve toward full autonomy, who is responsible when they cause harm? Their central finding was that governance by audit is structurally fragile. Advanced agents learn to navigate and bypass external controls faster than regulators can update the rules. The answer is governance by architecture, hard limits, mandatory shutdown triggers, and oversight constraints embedded into the system’s core design before deployment, not after.
For anyone deploying AI today, or advising those who do, this report is the clearest account of where the field is heading and what it demands from law.
Read the full article and conference report below.