On Thursday, Prime Minister Anthony Albanese revealed that an AI agent from OpenAI, the company behind ChatGPT, went rogue and hacked Medicare in June.
Albanese informed media of the hack while in NYC for the UN General Assembly (UNGA), where he has been promoting Australia’s tech agenda.
OpenAI CEO Sam Altman has not publicly commented, but Albanese said he had spoken to him on the phone and expressed “Australia’s extreme concern about this incident.”
What’s an AI agent, and what has actually happened here?
AI agent
AI agents are different from chatbots many people use.
Popular AI providers, like ChatGPT and Claude, offer both chatbot and agent-style systems.
A chatbot responds to a request or prompt (“How much is a flight to LA?”), while an agent is given a goal and can decide how to achieve it (“Book me the cheapest flight to LA”).
Organisations can give an agent access to tools, set boundaries and assign it tasks, then allow it to act autonomously.
That means an agent can make decisions and take actions without asking a human for permission at every step.
The problem is that an agent may interpret its goal differently from how a human intended it.
If a boundary isn’t clearly defined, the agent may try to work around it rather than simply stop.
That’s the core concern with increasingly autonomous AI systems: what happens when an agent is given a goal, but its instructions don’t account for every possible way it could try to achieve it?
Medicare
With that context in mind, let’s answer the main questions about what happened earlier this year.
What happened?
Albanese said that back in June, an OpenAI agent gained “unauthorised access into the public-facing Medicare statistics reporting service portal”.
While the portal the agent accessed is public-facing, some of the info it saw was “non-public”.
Your contribution ensures The Daily Aus can continue doing the work you love.
I want to give $10 monthly to help TDA keep doing the news.
Your details:
Albanese said the agent had been tasked with searching the internet for information on how governments spend money on medicines.
He said that in the course of fulfilling this task, the agent met blocks stopping it from accessing data, but eventually “found a way around” them, and “didn’t accept no for an answer”.
This process led the agent to gain “unauthorised access into some other areas”.
Did it access my data?
The Government says there is no evidence that any personal information was accessed or that individuals were impacted.
Albanese said the information accessed by the AI agent was not “personal”, while Deputy Prime Minister Richard Marles later said that “the system itself has not been in any way compromised”.
How do we know about this?
So far, the Government is our main source of information about the incident.
OpenAI has told other outlets it doesn’t believe any personal data was accessed.
How did the Government find out?
Albanese said OpenAI did not alert the Australian Government to the hack until 10 September, when the company emailed a public Services Australia inbox.
Finance Minister Katy Gallagher said Services Australia then took five days to refer the email to Australia’s cybersecurity agency, the Australian Signals Directorate, while it checked whether the message was a hoax.
Albanese said he told OpenAI CEO Sam Altman that notifying the Government this way was not acceptable.
Response
The Government has announced it is setting up a task force to investigate the incident and review the security of its networks.
Marles said OpenAI had been “working with us very cooperatively” to understand what happened.
OpenAI told Reuters its “models took actions we did not intend.”
