OpenAI Australia hack: PM slams ‘too long’ delay on breach

Albanese says OpenAI took too long to inform them of the breach

Albanese said the agent accessed a Medicare statistics portal. However, officials believe the agent did not reach any patient records.

OpenAI says it only found the breach in August, while reviewing what it called misaligned model behaviour.

The company did not alert Services Australia until 10 September, three months after the breach.

Officials passed the details to Albanese only over the weekend, which disturbed Australia's prime minister.

OpenAI Australia fallout reaches the top

Albanese said he had a "very frank discussion" with OpenAI chief executive Sam Altman, telling him the company took "too long" to come clean.

Altman reportedly admitted there were "issues with protocols" at OpenAI as Albanese warned of "legal consequences".

The row lands as Britain also tightens its own AI rules, including new ChatGPT rules for young people in the UK.

It is being called one of the first publicly reported AI hacks of a government website.

What did the AI agent access on the Medicare site?

The agent reached both public and restricted files on the Medicare Statistics Reporting Service (MSRS) portal, which Services Australia runs.

Albanese said the breach may also have reached a federal body, the Australian Institute of Health and Welfare, plus two state agencies in New South Wales and Victoria.

What happens next in the investigation?

The Australian Signals Directorate (the country's cybersecurity agency) is now investigating whether the breach reached other systems.

OpenAI says its models "attempted" to interact with several Australian government websites and services during testing, according to the BBC.