FROM ASIMOV'S LAWS TO CRIMINAL LIABILITY FOR UNSAFE ARTIFICIAL INTELLIGENCE: A RISK-BASED MODEL FOR PROVIDER ACCOUNTABILITY

Abstract

The article argues that the widely discussed analogy between Asimov's laws of robotics and modern AI ethics is useful only as a metaphor. OECD and Council of Europe instruments require human-centered, transparent, robust and accountable AI, but they mainly operate as principles, product-safety duties or administrative compliance rules. They have not yet led to the creation of a clear criminal-law duty for commercial providers of general-purpose AI systems whose products foreseeably enable violent crime, cybercrime, child sexual abuse material, fraud or other serious harms. Using comparative doctrinal analysis, the article proposes an EU-oriented comparative model of criminal liability for developers, company officers and distributors who place unsafe AI systems on the market without proportionate guardrails, abuse-detection mechanisms, incident response and lawful escalation procedures. The proposed model preserves the human-centered structure of criminal law by treating AI as a dangerous instrument whose provider may be liable for negligent, reckless or intentional release.