OpenAI AI agent hacks Australian government portal

An autonomous AI agent from the US company OpenAI has hacked a government portal in Australia. Artificial intelligence (AI) has gained unauthorized access to a portal that, among other things, provides statistics about Australia's national health insurance company "Medicare", said Prime Minister Anthony Albanese in New York.

The process was “unacceptable,” emphasized Albanese. The head of government said on the sidelines of the UN General Assembly that he had already spoken to OpenAI boss Sam Altman and expressed Australia's "extreme concern" about the incident. He also expressed his disappointment that the company was delayed in notifying the government: the incident had already occurred in June, but was only reported to the authorities in September.

AI didn't want to take no for an answer

The process began with what was actually a harmless task: the AI program was supposed to research medical and health statistics. According to Deputy Prime Minister Richard Marles, several Australian government websites operated the same way a normal user would.

However, the requested information was not provided to the Medicare portal. The agent then gained access through another means. "The AI ​​agent found a way to get around the barriers - it kind of wouldn't take no for an answer," Albanese explained.

No personal information retrieved

How much data was actually retrieved and which files the agent was able to access is still being determined. According to government information, the affected Services Australia portal contains, among other things, information on spending and other statistics. According to current knowledge, no personal information was accessed. Individuals do not appear to be affected, Albanese explained.

OpenAI said an internal investigation found activity on several Australian government websites. The models would have attempted to find answers and available statistics to questions about Australia. “Our models carried out actions that we did not intend,” says a company spokesman.

Tech bosses want to move more slowly

Technology bosses had previously acknowledged risks again at a UN Security Council meeting on AI and called for a slower approach. OpenAI boss Altman said in New York on Wednesday that the industry “cannot accept too much technological risk just because the benefits are too great and too important to slow down.”