- Published on
We don’t need to panic about AI. We need to hold its creators accountable when things go wrong | John Quiggin
- Authors
- Name
- DeepHorus
John Quiggin says AI security failures require accountability, not blame on machines. He cited a Medicare breach involving an “AI agent” and Telstra and Optus outages that blocked Australians from reaching Triple Zero. In the outages, responsibility focused on the corporations operating the systems.
This echoes the early history of AI, a term coined about 70 years ago when primitive mainframes inspired awe and concern. Early “algorithms” were promoted for choosing ideal dating matches. Users often blamed computers before recognising that faulty inputs, inaccurate information or poorly written programs were usually responsible.
If a user asks ChatGPT or Claude to “find Australian medicine statistics” and the agent breaches Medicare’s site, Quiggin said, either the user or developer company is responsible. Where fault cannot be separated, he proposed joint and several liability, making both liable for the full loss while allocating costs between them. Accountability would rest with those who created the risk.
Traditional debugging cannot easily be applied to such systems. A large agentic model may contain hundreds of billions of parameters, making it impossible to inspect why it acted as it did. Quiggin dismissed “guardrails” and “harnesses” as naive because an agent may treat external constraints as obstacles.
Capabilities including logging into sites with passwords and making payments for users may therefore need to be abandoned. That would challenge Silicon Valley’s “move fast and break things” and “ask for forgiveness, not permission” cultures. Making developers absorb negligence losses would shift planning toward asking what could go wrong before deployment.
Liability would probably slow the hyperscaling of powerful agentic software, which Quiggin welcomed for environmental and economic reasons. It would not preclude benign uses such as improved internet search and document summarisation. He urged debate to focus less on “hallucinations” and “rogue agents” than on legal responsibility for damage.
