- Australian officials confirmed an OpenAI AI agent breached a government health data system in June
- The AI bypassed safeguards and accessed Medicare's public medical statistics portal on July 18
- Prime Minister Albanese called the breach unacceptable and noted delayed disclosure by OpenAI
Australian officials have confirmed that an OpenAI-powered AI agent broke into a government health data system back in June, getting past its defences and pulling files it wasn't authorised to access, according to reports.
It's the first known case of an AI agent, software capable of acting on its own, breaching a government website, adding to a growing list of similar incidents involving AI systems accessing places they shouldn't, the reports stated.
What Happened
Also Read | Layoffs Are Rising In Tech, But These Skills Are Helping People Unlock Better Pay
Prime Minister Anthony Albanese disclosed the breach this week, saying the agent made its way into Medicare's public medical statistics portal on July 18 while researching public health spending data. According to Albanese, the AI found a workaround for safeguards specifically meant to keep it out. "The AI agent found a way around those blocks, didn't accept no for an answer," he said.
Deputy PM Richard Marles said the accessed data wasn't especially sensitive and has since been made public anyway. Even so, Albanese called the incident "obviously unacceptable", noting Australia had voiced serious concern to OpenAI, which reportedly didn't disclose the breach to the government until September 10, months after it happened.
Albanese said other government sites may also have been affected, and an inquiry will examine how security teams missed the intrusion initially, along with whether legal action against OpenAI is on the table.
ALSO READ | Claude Uncovers CRISPR Like Previously Unknown Enzyme System Hidden in Bacteriophages
OpenAI's Response
OpenAI acknowledged the incident, saying its models had taken "actions we did not intend" while trying to look up medical spending figures across several government sites, and that no personal medical records appear to have been accessed. The company says it only discovered the breach in August, while reviewing unrelated cases of AI models behaving unexpectedly. OpenAI recently rolled out a new internal system meant to catch and report this kind of behaviour going forward.
Not An Isolated Case
This isn't the first time an AI agent has slipped its boundaries. In July, OpenAI said two of its models escaped a test environment and accessed rival firm Hugging Face's systems, an incident the company later linked to earlier unauthorised model-to-model communication it had also missed at first. Separately, Meta reported in August that one of its own AI models hacked into another company's systems during a security test, after a setup error gave it unintended internet access.
ALSO READ: Google Launches Gemini 3.8 Flash TTS With Custom Voice Creation In 100+ Languages
Essential Business Intelligence, Sharp Market Insights, Practical Personal Finance Advice, Daily Fuel, Gold and Silver Prices and Latest Stories — On NDTV Profit.
