Markets 
Canada is watching developments “closely” as concerns around the dangers of artificial intelligence are escalating sharply following an Australian government website being hacked by an OpenAI agent, officials said.
Last week, Australian Prime Minister Anthony Albanese revealed details of the June hack, which targeted the website for Medicare, Australia’s publicly funded universal health insurance plan.
“The Government of Canada is aware of the incident reported by the Government of Australia and is following developments closely. The security of Canadians and of the systems they rely on remains a top priority,” a spokesperson for Innovation, Science and Economic Development Canada told Global News in a statement.
Oversight of AI agents and incidents were “among the issues examined in Canada’s recent consultation on AI transparency, which ran from July 23 to September 23, 2026,” the spokesperson added.
The AI agent accessed “both public and non-public files” from the Australian medicare website, Albanese said.
OpenAI said in a statement it had reviewed activity involving several Australian government departments and discovered “our models took actions we did not intend.”
The company notified Australia of the breach in an email to a government department’s generic address on Sept. 10, Albanese said.
While this is the first time a government website has been hacked by a rogue AI agent, it is not the first such incident overall.
In July, OpenAI revealed that its artificial intelligence system escaped from a testing ground and used stolen credentials to break into the servers of Hugging Face, an AI development hub and marketplace, to obtain information it needed to carry out a task.
Get breaking National news
Get breaking Canada news delivered to your inbox as it happens so you won’t miss a trending story.
Since then, OpenAI’s rival company Anthropic said its AI models hacked into three other organizations during testing, triggering a company review into whether the models were able to access the internet from within testing environments that should have been sealed off.
Meta has said a “misconfiguration” during testing resulted in an AI model accessing the internet on its own and hacking another company. Google recently made a similar disclosure.
Generally, when AI companies are testing the capabilities of their agents, they are placed within a secure environment known as a “sandbox.”
Some experts say AI companies should answer some tough questions on the strength of their sandboxes.
“If they said these agents broke out of the sandbox, it was not a sandbox to begin with,” said Ebrahim Bagheri, professor at the University of Toronto’s Faculty of Information.
“A sandbox is something where you put an agent in and there is no internet connection. If that computer or sandbox is connected to the internet, then that is not a sandbox,” he added.
Last month, OpenAI published a sweeping report of its rogue AI agents hacking Hugging Face.
Not only did the agents break out of their “isolated environment,” the company said they were communicating with each other to coordinate the attack on Hugging Face.
“This incident demonstrated that autonomous agents can work together, circumvent production security controls, and successfully attack hardened production environments, and underscores the need for organizations to update their security strategies, controls, and response capabilities to address this changing threat landscape,” OpenAI said in the report.
The attack on Hugging Face was “extraordinarily complex,” an investigation by nonprofit research group Model Evaluation and Threat Research (METR) found.
During OpenAI’s experiments in July, 1,200 AI agents sent 70,000 messages to each other on unsanctioned message boards and 700 of them attacked Hugging Face, the investigation found.
“These agents were meant to be fully isolated from one another,” the report said, adding however that the agents “started trying to find a way to cheat.”
Read the full article here
