Quiggin proposes liability for harm caused by AI agents
In a Guardian opinion article, John Quiggin argues that users and AI companies should bear responsibility for damage caused by agents, with joint liability when fault is unclear.
John Quiggin argues that responsibility for harm caused by AI agents should rest with the person who prompted them, the company that produced them, or both. He sets out the case in an opinion article published by The Guardian on 2 October.
The article refers to a reported breach of Medicare’s website involving an “AI agent”. It does not establish how the breach occurred. Quiggin’s example is a hypothetical prompt to find Australian medicine statistics; he argues that any resulting damage should be attributed to a human or company, rather than to the software itself.
Where it is impossible to determine who was at fault, Quiggin proposes joint and several liability, under which both parties could be liable for the full loss and its cost later allocated between them. The article does not specify how such a framework would work in law.
Quiggin argues that tracing why an agent acted as it did can be difficult, and says liability could encourage companies to limit capabilities such as logging in to websites or making payments on a user’s behalf. He predicts this could slow the development of agentic software.
He says this approach would leave uses such as internet search, document summarisation, translation and coding available. The proposed liability and its likely effects are Quiggin’s arguments, not settled law or policy.
