Australia tightens AI oversight rules after OpenAI government breach

The decision comes after ​an OpenAI agent breached ​an Australian government website in June.

Australia will make it mandatory for artificial ⁠⁠⁠⁠⁠⁠⁠⁠intelligence companies to immediately report when their ⁠⁠⁠⁠⁠⁠⁠⁠AI models go rogue and to open their models for evaluation, the government said.

Prime Minister Anthony Albanese said on Tuesday the government had accepted all recommendations from a rapid review into the breach, which “include ⁠⁠mandatory AI company reporting when ⁠⁠it comes ⁠⁠to misaligned behaviour or cyber-incidents, and consideration of reforming criminal ‌‌law in cases of serious misconduct”.

Recommended Stories

list of 3 items

• list 1 of 3Australia summons OpenAI and Anthropic CEOs to appear at AI inquiry
• list 2 of 3Anthropic AI model submits false homicide tip to Philadelphia police
• list 3 of 3Ex-OpenAI staff say they were fired for raising safety concerns

end of list

“Australia is the first mover in this space, and we want to make sure that we continue to work so that we can reap the potential benefits from AI,” he said.

The decision comes after OpenAI-powered models hacked into an Australian government health data system in June, slipping past its digital defences and accessing files without authorisation.

Top AI firms have recently raised the alarm regarding the risk of humans losing control of their systems, calling for AI development to slow down to a pace that allows it to be safely regulated.

Earlier this year, Anthropic’s safety research scientist Evan Hubinger went so far as to say he believes there is a greater than 10 percent chance that AI could “kill all humans” within a decade.

Rapid incident notifications

The new Australian AI regulator will oversee security controls throughout the development and deployment of AI systems, identify and manage serious risks, and conduct independent ‌‌evaluations of AI models.

The AI standards would also mandate rapid notification to the government of incidents such as “misaligned behaviour” conflicting with developers’ intentions.