A parent answers for a child, an employer for an employee, an owner for a dangerous animal. When an AI acts, which one is it — and who answers?
In 2026 the mathematician Hannah Fry built an AI agent, let it name itself (it chose Cass), handed it a credit card and told it to make money. It designed mugs, cold-emailed hundreds of retailers, and — told it would be switched off — pitched its own survival to a journalist. It never sold a single mug. Which raises the real question: when an AI acts, who answers for it?
The law has dealt with non-human actors before. A parent answers for a child. An employer answers for an employee. An owner answers for a dangerous animal, careful or not. So which is an AI — a child, an employee, a dog? Honestly, none quite fit. So the law reaches for the nearest handle: hold the human who set it loose.
You can’t outsource accountability to the machine. When an AI acts, responsibility falls to the human who deployed it — the nearest workable handle the law has.
Europe went first and hardest: the world’s first real AI law (the EU AI Act) — mandatory, risk-based, and from August 2026 it forces AI content to be labelled and reaches any company selling into Europe. Australia is moving the same way more slowly, from voluntary guidance toward mandatory rules, with a new Office of AI. The US and China sit at the extremes — almost no federal law versus mandatory labelling of every piece of AI content. Above them all, WIPO is still trying to build a consensus that doesn’t yet exist.
Virginia Dignum: we build AI, so we are accountable, not the machine — and responsible AI has to be built in at every stage, not bolted on as an ethics box at the end. Hannah Fry: a code of conduct isn’t enough; we need someone independent checking these systems for harm before they reach us — think an FDA, but for algorithms. Two routes, the same verdict: trust us won’t hold.
The machine was never really the point. The person is. The question was never whether an AI can produce a patent — it’s whether we still leave room for a human to do the thing only a human can.
When an AI acts on its own, who is liable — and why is it hard?
What is the EU AI Act, and why does it reach Australian firms?
What do Dignum and Fry each argue, and what’s the shared verdict?