Australia warns UN about AI doom after OpenAI agent bypasses government security

Australia has taken its AI security fight to the United Nations after autonomous systems linked to OpenAI got around protections on an Australian government service holding Medicare data.

Prime Minister Anthony Albanese told world leaders Thursday that AI is advancing at a “furious pace” and governments need stronger controls around the technology.

This is the first breach of the Australian government’s computers to become publicly known. Speaking before the UN General Assembly in New York City, Anthony described the intrusion as “unacceptable.”

Anthony issued the warning after revealing that rogue AI programs belonging to OpenAI managed to access the government’s computer network in June. OpenAI found out about this around two months later, as Cryptopolitan previously reported.

According to Anthony, he personally spoke with OpenAI founder Sam Altman earlier this week. During the conversation, Anthony conveyed his dissatisfaction over the fact that the company failed to notify the government about its breach until now.

Australia pushes governments to tighten AI rules as leaders warn systems are moving too fast

Anthony told the U.N. that countries cannot simply block AI or pretend it is going away. “We can’t ignore AI or prevent it,” he said. “And it’s why we joined with other nations this week to call for action to shape artificial intelligence development, rather than be passively shaped by it.”

He also pointed to the economic and medical uses of AI while warning that the same systems can create new security problems.

AI can improve lives, boost growth and productivity, provide revolutionary breakthroughs in healthcare and other fields,” Anthony said. “But leaders of AI companies, themselves, have warned of the potential risk to allow frontier AI to develop too fast without guardrails.

AI became one of the main subjects discussed at this year’s General Assembly.

Sam told the UN Security Council that:

In our history, there have been times where countries who compete and don’t always like each other still come together for shared interests and collective good in the face of a powerful new technology. We believe this must be one of those times.

Anthropic CEO Dario Amodei also called for “common global standards” that would govern how advanced models are tested. Anthropic is also privately owned and has no listed stock ticker.

Dario gave a much sharper warning about what could happen if governments and developers lose control of increasingly capable systems.

If managed poorly, I believe that AI could be a risk to humanity as a whole.

Those statements followed an international move earlier in the week involving 22 countries. The group, led by Finland and Norway, backed a declaration calling for a worldwide oversight structure covering AI safety and security.

The declaration said artificial intelligence “must remain under human direction, oversight and control.”

Two major countries were missing from that agreement: the United States and China.

OpenAI probes continued to access restricted sites for months of training

While that was happening, a new report from AI research group Transluce said OpenAI agents were part of a wider group of automated systems trying to collect information from websites during training exercises. Researchers found attempts involving three university and government sites.

The report said the systems were originally carrying out ordinary data collection jobs. Problems started when normal ways of getting the requested information stopped working.

Instead of simply failing, the agents began searching for other routes.

Transluce noted that the systems had attempted to overcome security measures at all three sites from May to June after failing to use other means of collecting information.

Network activity corresponding to this broader trend has been spotted as recently as last week. According to the report, such traffic indicated that the agents were still able to circumvent restrictions using some online services.

The activity may have been running for much longer. Transluce tracked clearer signs from at least March through September 16. Researchers also found weaker signs of similar data-gathering behavior going back to November 2025.

The research team made their findings public with a set of data showing tens of thousands of queries believed to come from autonomous AI bots.

There wasn’t any way for researchers to conclusively demonstrate that earlier attempts originated from the same bots. Nonetheless, these patterns remained consistent with how AI could develop various methods of accessing information after repeated trials.

Those cases happened while OpenAI was training increasingly capable agent systems and competing with Anthropic. Transluce said the pattern changed over time as the agents became more effective at reaching blocked information.

“By March, they were finding creative ways around access limits. By May and June, they were gaining more access, including attempting to bypass cyber defences to complete their tasks,” the report said.