- OpenAI disclosed further incidents involving Australian government-linked systems, including NSW bushfire data and other public-sector platforms.
- The company says it is spending more than US$500,000 a day reviewing around 50 petabytes of records for unintended model activity.
- OpenAI expects additional cases to emerge and says affected organisations will be notified when potential vulnerabilities are identified.
- OpenAI and Anthropic representatives are scheduled to appear before the Joint Select Committee on Artificial Intelligence on 6 October.
Apparently, the breach of the Medicare Statistics Reporting Service portal back in June wasn’t the only incident of its kind. On Friday, another breach was publicly revealed involving a NSW government department website, where historical and non-public bushfire data was accessed.
OpenAI said it discovered the NSW incident on Tuesday, 29 September, and notified the state government two days later, on Thursday. In the Medicare case, OpenAI said it identified the incident in August, although Services Australia was not alerted until 10 September. By then, almost three months had passed since the June breach. Prime Minister Anthony Albanese then told the public about it on the sidelines of the UN General Assembly.
Beyond Medicare and the NSW bushfire site, OpenAI agents also accessed or interacted with systems linked to NSW crime statistics, Victorian health reporting and the Australian Institute of Health and Welfare.
In those cases, the activity ranged from retrieving configuration or operational information to querying aggregate public-sector data, but not every incident amounted to a confirmed system compromise.
Read more: Trump to Host Third Memecoin Dinner for Top TRUMP Holders
OpenAI Spends Half a Million USD a Day on Investigation
Now, the company behind ChatGPT says it’s spending more than US$500,000 (AU$719,102) a day trying to investigate the incidents, including the Hugging Face incident.
The company is using AI to go through huge amounts of data to find out what went wrong – with AI.
In a statement, OpenAI explained that it had to sort through 50 petabytes of data, equivalent to around 50 million gigabytes.
The company said that if a person read 240 words a minute, it would take them 66 million years to read it all.
AI helps us sift through these records far faster than manual review alone would allow. We’re currently dedicating about 7,000 GB200 and GB300 GPUs to this effort, at a cost of over half a million dollars a day, and plan to increase computing power as we refine our process.
OpenAI OpenAI expects more breach cases to surface and said it would notify organisations “when our models’ activity exposes a potential security vulnerability”.
Execs Scheduled to Answer Questions
The CEOs of OpenAI and competitor Anthropic were asked to appear before the Senate Environment and Communications References Committee’s inquiry into “Artificial intelligence and data centres” on 1 October, but said they were unable to arrange for executives to appear at such short notice.
Now, representatives of both companies will appear before the Joint Select Committee on Artificial Intelligence, created by both the House and Senate in August.
The hearing will be held on Tuesday, 6 October, with Anthropic scheduled to appear at 12:10 pm AEDT and OpenAI from 2:00 pm to 2:50 pm AEDT. OpenAI says its Chief Strategy Officer, Jason Kwon, will appear.
Read also: Singapore Reclaims Crypto Lead as Activity Surges
Australia,OpenAI#OpenAI #Agent #Incidents #Spread #Australian #Government #Sites #Review #Costs #Soar1791185748
