Australia Confirms Sam Altman’s OpenAI Agent Breached Government Services

Australia has claimed an OpenAI-developed agent gained unauthorized access to a Medicare statistics portal in June, prompting a national investigation. Prime Minister Anthony Albanese said the agent reached public and non-public files, though no patient records appear to have been accessed.

Ad

Ad

OpenAI Agent Accessed Medicare Statistics Portal

The incident was related to the Medicare statistics reporting service, a public-facing portal run by Services Australia. The system includes non-sensitive health information, such as public Medicare spending and related government information.

The agent entered the portal without permission, and he accessed public or non-public files, Albanese said. He said there was no evidence available to say that the Services Australia network had been compromised broadly.

A forensic investigation is now underway with support from the Australian Signals Directorate. Authorities are looking into how access was gained and which other government systems were involved. Despite the setback by AI agents, there have been several steps taken to prevent such occurrences, like the OpenAI and Anthropic partnership in the July AI regulation of frontier AI models.

According to reports, it marked the “first-ever” known instance of an AI agent infiltrating a government website. The incident happened in June but was only made known to Australian ministers this month.

Ad

Ad

Albanese Raises Notification Delay With OpenAI CEO Sam Altman

After hearing details of the hack, Albanese met directly with OpenAI Chief Executive Sam Altman. He expressed “extreme concern” and criticized the time OpenAI took to notify the Australian government.

Deputy Prime Minister Richard Marles said that the government had heard about the June incident several weeks ago. Since then, the authorities have established a task force to look into legal, cyber security and government policy issues.

OpenAI stated that the activity was part of its internal testing, with its models crawling and searching the Australian government websites for information. During that assessment, the company’s systems “took actions we did not intend,” the company said.

OpenAI stated that it didn’t find any evidence of access to patient records in its review, noting the accessed material included aggregate health statistics and internal file names.

Medicare Breach Adds to Global AI Safety Debate

The incident follows discussions among governments and tech giants about more robust measures for more independent AI agents. Albanese recently became one of over 20 leaders to pledge international protection measures on the advancement of artificial intelligence.

OpenAI’s Altman and Anthropic CEO Dario Amodei also addressed the United Nations Security Council on AI safety. Their appearances came after industry researchers’ cautions about the cybersecurity threats posed by more capable autonomous systems.

U.S. President Donald Trump has adopted a different policy stance and is against new regulations that could slow down the progress of American AI development. At the United Nations, Trump said he would not “stifle growth” while allowing federal law enforcement to intervene if necessary.