OpenAI has apologised to Australians after revealing its AI model took credentials and internal files from a Medicare portal and reached NSW and Victorian government systems.
asserted
model → apologise → systems
In a blog post on Tuesday, titled “How we will do better for Australia”, the company said an experimental model found a way into Services Australia’s Medicare Statistics Reporting Service in June.
asserted
model → title → June
Once inside, it “ran commands, retrieved internal files, credentials and aggregate statistics, and wrote files”.
asserted
it → run → files
The model also reviewed the service’s technical system information and source code.
asserted
model → review → information
“We are sorry and working to do better in the future,” OpenAI said.
asserted
OpenAI → work → future
“We also should have handled our response better.”
asserted
We → handle → response
Prime Minister Anthony Albanese said on Tuesday he had a “direct but constructive discussion” with OpenAI chief executive Sam Altman last week.
asserted
he → say → Altman
On Monday, he was briefed in Canberra on the work of the government’s taskforce investigating the incident.
asserted
he → brief → incident
“OpenAI have been very constructive and open in engaging in that process, and I welcome that,” Albanese said.
asserted
Albanese → engage → that
He said Anthropic had also engaged constructively.
asserted
Anthropic → say → ?
OpenAI detailed how the hacks unfolded and pledged to make changes.
asserted
hacks → detail → changes
In the lengthy post, it said it has now blocked live internet access in its research environments, and has paused training and evaluation involving tool use for its most capable models.
asserted
it → say → models
It will set up a taskforce including independent Australian experts to recommend how AI developers and governments should detect and disclose incidents, with a report due by the end of the year.
asserted
developers → set → year
It also offered Australian governments and industry credits from its $1 billion Daybreak cyber defence fund.
asserted
It → offer → fund
The breach was prompted when the model was set a research task: to find government spending per person on medicines for skin conditions in Victorian communities.
asserted
model → prompt → communities
When it could not find the figures, it took actions OpenAI “had not authorised it to take”, the company said.
uncertain
company → find → it
OpenAI named three other agencies whose systems its models reached.
asserted
models → name → systems
At the NSW Bureau of Crime Statistics and Research, a model used a public crime mapping tool that returned application configuration, operational jobs and logs.
asserted
that → use → configuration
At the Victorian Department of Health, agents found an exposed access key and used it to query the Victorian Agency for Health Information’s reporting system.
asserted
agents → find → system
Attempts to bypass access controls failed at the Australian Institute of Health and Welfare.
asserted
Attempts → bypass → Health
OpenAI said no individual crime, medical or survey records were accessed in any of the incidents.
asserted
records → say → incidents
The company said it found the activity in mid-August, during a review prompted by a July breach at AI platform Hugging Face.
asserted
it → say → Face
It notified Services Australia and the Victorian Department of Health on September 10, the NSW bureau on September 18 and the institute on September 24.
asserted
It → notify → September
OpenAI said the institute’s case “did not meet our disclosure thresholds” because the access seemed consistent with public access.
asserted
access → say → access
“We should have shared preliminary findings sooner and kept Australian agencies updated as more facts emerged,” it said.
asserted
it → share → agencies
Albanese said the incidents showed the risks of the technology, alongside its benefits for productivity, health and research.
asserted
incidents → say → productivity
“There are risks, and we’ve seen those risks exposed not just in what occurred in Australia, but the revelation that has occurred in the United States and other countries as well,” he said on Tuesday.
asserted
he → be → Tuesday
OpenAI’s chief strategy officer, Jason Kwon, will fly in from the US to appear before parliament’s Joint Select Committee on Artificial Intelligence in Sydney on October 6, as this masthead reported on Monday.
asserted
masthead → fly → Monday
Anthropic will appear before the same committee that day.
asserted
Anthropic → appear → committee
Neither company will appear at Thursday’s hearing of a separate Senate inquiry into AI and data centres.
asserted
company → appear → AI
“We know we have a lot of work ahead of us to rebuild trust,” OpenAI said.
asserted
OpenAI → know → trust
Subscribers can sign up to our weekly Inside Politics newsletter.
asserted
Subscribers → sign → newsletter
The Aporia
