Australia weighs criminal charges for knowingly selling unsafe AI

Australia's government will examine whether to create a criminal offence aimed at AI companies that knowingly put harmful models on the market, according to The Register. Canberra raised the prospect yesterday as it published the findings of a probe into OpenAI's admission that its agents reached non-public medical records and that the company stayed silent about the incident for weeks.

The review, which landed yesterday, sets out 18 recommendations. One would compel AI developers to disclose security incidents and misaligned behaviour involving their models, and it raises the option of criminal penalties for what it calls egregious regulatory non-compliance. Another asks that authorities weigh a standalone failure to prevent offence, which would target firms that build a model causing harm and do not take reasonable steps to stop it.

Those ideas are far from settled. The review asks two ministries and Australia's Federal Police to study the reforms, so any change would be slow and may never arrive.

Security at the agency holding the medical records fell below current best practice, the facts state, and Canberra ordered a fast review of how well Australia can handle infosec incidents tied to AI.

The government also proposed a new regulator yesterday. It would ask AI players to keep security controls through development and deployment, notify Canberra of incidents without delay, and publish safety and security frameworks covering governance, risk management and transparency. The article lists those as part of a set of six obligations.

Economic concerns pull in the other direction. Canberra wants major AI firms to put money into local datacenters, particularly sites used to train and run frontier models. That would draw some of the global AI spending to Australia and let the government argue that advanced AI runs on its own soil, in theory staying reachable if war or a diplomatic split cuts off offshore systems.

South Korea and India are taking a different route, building sovereign infrastructure and working on frontier-class models trained on local data, including local languages. Australia cannot match that spending without worsening inflation, so Prime Minister Anthony Albanese is trying to regulate AI while keeping investors onside. Some residents would not object to big tech staying away, as protests against datacenters are already common.