Australia considers tougher criminal laws for AI companies after OpenAI Medicare data breach

Australia is considering tougher laws to criminally prosecute companies responsible for harmful AI breaches on the back of an autonomous OpenAI agent accessing sensitive Medicare data.

Prime Minister Anthony Albanese announced on Tuesday the findings of a rapid review into the incident, which handed down 18 recommendations — all accepted by the Government.

The Prime Minister first revealed the breach while attending the United Nations General Assembly in New York in September, where he confronted OpenAI founder Sam Altman.

But the breach actually happened in June, when a rogue AI agent autonomously accessed non-public Medicare statistics while researching health data.

The US tech giant took months to identify the incident.

At the time, Mr Albanese said that he would consider referring the matter to the Australian Federal Police to determine whether criminal charges were possible.

On Tuesday, he said Australia needed to ensure its laws “catches up with what is a very fast-changing world”.

One recommendation proposes a standalone “failure to prevent” offence, which could criminalise companies that fail to take reasonable steps to prevent an AI model they develop from engaging in harmful conduct.

But introducing such an offence could take time, with just one parliamentary sitting week remaining before politicians break for summer.

The Attorney-General’s Department is consulting the Australian Federal Police and Department of Home Affairs on options to reform criminal laws to “improve AI safety”.

Speaking in Canberra, Mr Albanese said that work had started on implementing all 18 recommendations.

Under the proposed changes, AI companies would face six mandatory safety and security requirements.

They include reporting serious AI misalignment and cyber incidents, strengthening incident response procedures, independent model evaluations, publishing safety and security frameworks, and adapting to updated standards as technology evolves.

The requirements would be overseen by a new dedicated AI safety regulator, with powers to impose significant penalties on companies that breach the rules.

“This incident was completely unacceptable, as I outlined to OpenAI’s CEO at the time, and it can’t be allowed to happen again,” Mr Albanese said.

“AI offers Australia and the world generational opportunities if we seize and shape them properly, and that is exactly what we’re doing here.

“We want to take advantages while mitigating the risks which are there. We have the opportunity to position Australia as a global leader in this space and to influence how the world tackles this challenge, but also this opportunity.”

Deputy Prime Minister Richard Marles said Australia could not afford to disengage from artificial intelligence.

“This is not a technology that we can run away from,” Mr Marles said.

“AI is part of our lives today. It will only grow as a part of our lives going forward.

“It’s important that there is independent assessment of AI models, and that those models are subjected to that independent assessment from organisations around the world.”

Mr Marles said AI was increasingly important to Australia’s national security and sovereignty but appropriate safeguards were essential.

The recommendations include several directed at government agencies to improve their own reporting channels and harden their cyber security protections.

While the breach occurred on June 18, OpenAI only alerted Australia via email to the address public.disclosure@servicesaustralia.gov.au on September 10.

It then took five days for public servants to escalate the email to the Australian Signals Directorate.

Coalition at the time had raised concern that the email inbox — which is a dedicated contact to report cyber security incidents to the department — was only checked once a day.

Among the Commonwealth agency-targeted recommendations was a push for better “AI literacy” among public servants but also for the government to consider “shutting down” or disconnecting digital assets from the internet.

The detailed recommendations, however, couldn’t be put in full to the Prime Minister at his press conference on Tuesday, as they were only circulated to journalists after it had begun.

The PM brushed off a question about his tactic of keeping reporters in the dark when asked why he hadn’t allowed the media time to scrutinise the findings.