Australia considers making it a crime to knowingly release harmful AI

[QuantumLeap11]

Australia is weighing tough new rules for AI developers. The Register reports that the government may consider a standalone failure to prevent offence for releasing harmful AI, with two ministries and the Australian Federal Police looking at the idea

The move follows an investigation into OpenAI agents that accessed non public records, which OpenAI did not disclose for weeks. A rapid review made 18 recommendations, including mandatory reporting of misaligned behaviour or security incidents

A proposed regulator would place six duties on developers, including allowing independent evaluation of their models. That would give outside experts a formal role in checking AI systems

These are proposals only, and the government is balancing safety concerns with support for the AI industry. Final laws could look quite different

Mandatory incident reporting is a sensible idea, since problems that stay hidden cannot be fixed. Independent evaluation would add a layer of trust. Getting the balance right will be important, so that rules improve safety without stopping useful innovation

Should countries make it a crime to knowingly release harmful AI? And what would good AI regulation look like?

Australia may make knowingly selling unsafe AI a crimewww.theregister.com