How far can responsibility for AI take us—and when must we consider responsibility of AI itself? 🤔 My timeline is full of “rogue AI”, increasingly disturbing details emerging about AI incidents (esp. the Hugging Face incident), new safety initiatives across industry and independent research, and growing calls from politicians and the public for tighter regulation... All of this raises a question: is existing law still enough? In my view, existing criminal law has so far provided a framework for addressing much of the socially harmful conduct associated with AI. But the Hugging Face incident is a wrecking ball to the comfortable assumption that these frameworks will always suffice... In my research, I distinguish at least four roles AI can play in crime: 1️⃣ a tool used to commit an intentional offence; 2️⃣ a cause of harm potentially attributable to someone’s criminal negligence; 3️⃣ a target of an attack; 4️⃣ a “quasi-perpetrator”, in cases where no human or legal person can be held criminally liable. These roles help frame three models of criminal liability: 1️⃣ Criminal liability of individuals and legal persons for using AI as an instrument of crime. 2️⃣ Criminal liability of persons for failing to exercise due care when developing, deploying or using AI. 3️⃣ A proposed model of direct criminal liability of AI itself, where the first two models prove insufficient. For the third, I now favour a narrowly defined, functional legal personality of AI agents, encompassing a legally distinct pool of assets and the possibility of criminal sanctions. Human and corporate liability must remain primary, and any such framework must preserve that. The question is what legal status and sanctions would allow criminal law to fulfil its functions as AI agents become more autonomous and adaptable. Simply tightening rules built around human actors may leave that question unanswered, leaving us behind what I call the “anthropomorphic wall”. I have a paper on these issues coming out in the Czech journal Trestní právo soon, and I am already preparing another. The detailed arguments will follow in those papers, but I think the discussion needs to happen now. If law wants to retain its ambition to regulate reality, it must first understand that reality—and be willing to reconsider assumptions built into its categories. Where would you draw the boundary between responsibility for AI and responsibility of AI itself? And what should follow when no human or legal person can be held criminally liable? #AIGovernance #AICrime #RogueAI