When an AI bot causes harm, who pays? Australian legal experts point to the humans who deployed it

Key points

• AI agents, software that carries out multi-step tasks autonomously, cannot be held legally responsible for harm they cause under current law.
• Australia has recorded what experts describe as its first reported automated hacking incident involving an AI agent.
• Professor Jeannie Paterson says deploying an AI agent that causes foreseeable harm makes the deployer legally responsible, even without intent.
• Developer liability remains an open question, and courts have not yet tested it.

An AI agent caused harm in Australia, and nobody knows exactly who should pay. That uncomfortable question is now in front of legal scholars and businesses that have started handing autonomous software to customers and staff.

What actually happened?

Australia recorded what experts are calling its first reported automated hacking incident involving an AI agent. The Guardian AI reported the case, though the specific target and operator have not been publicly named. A piece of software acted, caused damage, and had no legal identity of its own.

AI agents differ from ordinary chatbots. A chatbot answers questions; an agent takes actions, browsing the web, sending emails, writing and running code, sometimes without asking permission at each step. That autonomy is precisely what makes assigning blame complicated. We first covered the legal exposure of autonomous AI on 21 July 2026, and the liability question has sharpened with every incident since.

So who is legally responsible?

The human or company that deployed the agent. Full stop.

"If I deploy an AI agent and it causes harm to someone else, I am responsible for that harm," says Professor Jeannie Paterson, a legal expert on AI accountability. "Even if I didn't intend for that to happen, it was foreseeable, and I should be taking responsibility."

The logic mirrors how existing law handles faulty equipment: if a company's machinery injures someone, the company answers for it. The machine does not. AI agents occupy the same legal category for now.

Whether liability can climb further up the chain, to the developers who built the agent, depends on whether harm traces back to a design flaw rather than how the agent was used. That argument hasn't been tested in Australian courts, or most others.

What does this mean for ordinary people?

If a business deploys an AI agent that acts on your data, your accounts, or your devices and something goes wrong, that business is your first port of call, not the AI company whose technology sits underneath it.

For any business considering deployment: the speed these tools offer comes with direct legal exposure. An agent that sends emails, books appointments, or touches third-party systems on your behalf is, legally, your action. The debate over who writes the rules is still live, which means operators are carrying risk that legislation hasn't caught up with.

The plain judgement: the deployer-pays principle is cleaner than it sounds in practice. Real harm involves messy chains of decisions, from the lab that trained the model to the business that configured it to the user who set it loose. Courts haven't had to untangle that chain yet. When they do, the developer question will stop being theoretical.

Common questions

Can the AI company itself be sued?

Not straightforwardly, at least not yet. Current legal thinking places responsibility on the person or business that chose to deploy the agent, though developer liability remains an open question as more incidents occur.

Should I be worried about AI agents businesses use on my behalf?

Ask any company what automated tools act on your data and what their complaints process is if something goes wrong. Your consumer rights are the same as with any other service.