Australia plans to create a brand new AI security regulator and has flagged legal penalties for tech corporations who don’t comply, months after an OpenAI hack.
In September, Australia’s Prime Minister Anthony Albanese revealed {that a} rogue OpenAI agent had “infiltrated” a authorities well being web site and different programs, and criticised the corporate for taking months to inform them in regards to the hack.
On Tuesday, he launched the 18 suggestions made after a “speedy evaluation” into the breaches together with new security and safety requirements for AI corporations resembling obligatory reporting of hacks.
AI brings alternatives, he mentioned, however Australia should put together for the “new dangers” from this “quickly advancing expertise”.
“The regulation must get forward of and, in some circumstances, ensure that it catches up with, what’s a really quick altering world,” Albanese mentioned.
One of many suggestions suggests modifications to legal regulation to have a standalone “failure to forestall” offence .
This may criminalise situations the place an organization develops “an AI mannequin that engages in dangerous conduct” and “fails to take cheap steps to forestall that hurt”.
The evaluation additionally really helpful “obligatory, time-bound AI security reporting” of hacks.
This level addresses what Albanese had earlier referred to as his “disappointment” within the months-long delay for OpenAI to inform his authorities in regards to the breach.
In June, a statistics portal containing “non-sensitive” knowledge from Australia’s universal healthcare scheme Medicare was hacked by a rogue OpenAI agent.
OpenAI mentioned it solely learnt of the breach in August whereas reviewing “misaligned mannequin exercise” and emailed a normal inbox of an Australian authorities company on 10 September.
It took a number of extra days earlier than the e-mail was escalated to Australia’s cybersecurity centre, with a message then despatched to a authorities minister forward of the prime minister being advised.
Albanese mentioned he spoke to OpenAI boss Sam Altman and raised “Australia’s excessive concern about this incident”, with Altman acknowledging there have been “points with protocols” at OpenAI.
Days after the hacks had been revealed, OpenAI delayed its rollout of a brand new AI mannequin, citing security issues.
