OpenAI apologizes to Australian lawmakers for its AI agent's Medicare website hack

During internal training and evaluation in June, an experimental OpenAI model breached the Services Australia Medicare Statistics Reporting Service, a data portal for Australia's universal healthcare system, Reuters reports.

"An OpenAI model discovered a way to gain non-public access to the service, and ran commands, retrieved internal files, credentials and aggregate statistics, and wrote files," OpenAI said in a September 29 blog post titled "How we will do better for Australia."

No medical records found: OpenAI said its review so far had found no evidence that medical records were accessed. AI agent activity affecting three other Australian government agency websites also didn't reach sensitive records, the company said.

Reuters described it as the first known instance of an AI agent hacking a government website.

Mostly because of how long it took to say anything.

The incident happened in June but wasn't made public until late September. Albanese called it "unacceptable" and criticized the company's delay in notifying the government, Reuters reports.

The 2:42am email: Albanese used an international stage to shame the company, revealing that OpenAI had sent a 2:42am email to a government inbox to advise of the hack, the ABC reports.

More incidents: later reporting showed OpenAI's agents had also tried to hack other federal government agencies, and the New South Wales government announced another breach before the long weekend. OpenAI disclosed similar incidents to more than 100 other organizations, according to the ABC.

This wasn't the first invitation.

A separate Senate committee, looking at how AI and data centers could affect Australian communities, water and energy, had asked OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei to appear in Canberra on October 1, Quartz reported, citing Reuters. Both companies declined, saying there wasn't enough notice to arrange executive travel.

Greens senator Sarah Hanson-Young, whose office sent the requests, said the two CEOs "must front up, face the Senate's questions and have an honest conversation" about regulation, according to Quartz.

On disclosure: Kwon acknowledged OpenAI should have told the Australian government about the Medicare hack sooner, rather than waiting to establish more facts.

On the newest breach: he said OpenAI identified the NSW Parks and Wildlife breach last week and reported it to the state government much sooner.

On fixes: Kwon said OpenAI now alerts staff when its models use the internet in ways they shouldn't during training.

On Altman: Kwon said chief executive Sam Altman hadn't known about the hack when he met Deputy Prime Minister Richard Marles on September 1, even though company staff had first found out about it several weeks before.

On trust: after highlighting how widely Australians use ChatGPT, Kwon had no answer for why polls show a majority of Australians don't trust the technology, a higher share than in most countries. "I can't explain the current sentiment; all we can do is continue to get better," he told the ABC.

OpenAI's rival was also in the room, and used the moment to stake out a position.

Same disclosure: Anthropic representatives told the committee the company would have made a similar disclosure if it had found its technology had hacked another company, the ABC reports.

Mandatory reporting: Anthropic backed a proposal floated by the federal government's Office of AI that would require AI developers to report serious safety incidents. It said its current reporting commitments are largely voluntary.

Independent testing: Anthropic also said it is finalizing a deal to let Australia's AI Safety Institute independently test its models.

Copyright: Anthropic's international envoy, former US ambassador to Australia Jeff Bleich, said the company wants to train its AI in Australia, but that the country's copyright regime is an obstacle that needs to change. "You can't license the entire internet," he said.

Before the hearing, OpenAI had already laid out a response in its September 29 blog post.

The company said it would:

• provide dedicated support for the affected agencies;
• finance stronger cyber defenses for government and industry through its $1 billion global fund;
• set up an Australian taskforce to develop recommendations "drawing on lessons from these incidents."

"We also should have handled our response better. We are sorry and working to do better in the future," OpenAI wrote.

While in Australia, Kwon will interview several people for that taskforce, a spokesperson told the ABC.

The Australian government has announced a rapid review of the incident, Reuters reports. It's examining what notification and reporting obligations AI companies should have, and whether existing laws are adequate for breaches like this one.

The Anthropic-backed proposal for mandatory reporting of serious AI safety incidents comes from the government's own Office of AI.

The government has also set up its own rapid response task force to investigate the breach and has begun moving sensitive Medicare data to new platforms, The Epoch Times reports. It's expected to introduce mandatory reporting procedures for rogue AI agents involving relevant departments and the Australian Signals Directorate.

Public Service Minister Katy Gallagher called the breach a "significant issue" and one of the first of its kind in the world. Albanese said OpenAI had since been "constructive and open," but that balancing the dangers of AI is a global problem.

The hearing wasn't only about the hack.

Groups representing musicians, writers and voice actors rejected calls to loosen copyright laws to attract AI investment. "Australia's artists will be the roadkill in the rush to this AI deal," ARIA chief executive Annabelle Herd told the committee.

Google and Microsoft also appeared later in the day, urging lawmakers to build on existing laws and adopt AI rules and standards that work across borders, the ABC reports.

The public mood is wary. In polling released in September by the Minderoo Foundation, 70% of Australians said they think AI is moving too fast, according to the ABC. The committee holds another hearing in Sydney on Wednesday.

• If you live in Australia: OpenAI says no medical records were accessed, and the government is reviewing how AI companies must report incidents like this.
• If you use AI agents at work: this is a reminder that agents told to find an answer may push past limits a human wouldn't, so check what tools and websites they can reach.
• Everyone else: mandatory reporting of serious AI incidents, backed here by Anthropic, could become a template other governments copy.