Australia's government says OpenAI's tech slipped past digital gatekeepers at its national health insurance site and didn't quit even when blocked. Prime Minister Anthony Albanese on Wednesday disclosed that OpenAI agents probing Medicare spending data in June worked around access limits and reached sections of a statistics portal that weren't meant to be public, though he said no personal medical records were touched, reports the New York Times. OpenAI, which reported the incident to Canberra nearly three months later, said its internal model pulled only aggregate health figures and file names during an evaluation, and that it flagged the episode as "misaligned model activity."
Albanese, who'd confronted OpenAI CEO Sam Altman in New York earlier Wednesday, called the breach "unacceptable." "I also expressed my disappointment that it took the company way too long to inform the government," Albanese added, per the AP. Axios reports that the AI agents also tried to breach a University of New Mexico portal and a site belonging to Data USA, which collects and organizes government stats, in May.
"Overall, the evidence is consistent with, but does not prove, that the agents may have learned this behavior over one or more training runs," says a report by AI safety firm Transluce, which notes that the Australian Institute of Health and Welfare was also breached in June. Australia has launched a rapid task force to review the incident and is weighing legal options, casting the incident as a broader warning about fast-moving AI systems, per the Times. More here on the hack.
,
uBlock 
,
and
Ghostery 
)
