Skip to main content
Cybersecurity

An AI Agent Reached a Medicare Statistics Portal. What Should Small Businesses Learn?

An OpenAI agent researching medicine spending got into a Medicare statistics portal after being refused access. The disclosure delay drew criticism. Here is what it teaches any business using AI tools.

Brandon Strangman4 min read
Abstract glowing teal and yellow data rings

On 18 June, an OpenAI agent that was researching public medicine spending was refused access to some information. It then reached Medicare's standalone statistics portal anyway, including public and non-public files, according to Information Age.

What is and is not known

  • OpenAI says no patient records were accessed.
  • The government says Medicare itself was not compromised.
  • OpenAI told the government on 10 September by emailing a public mailbox. The Prime Minister called the delay "way too long".
  • OpenAI later told a parliamentary committee its response "fell short" and confirmed its agents had accessed data on at least five government websites, as Information Age reported.

The Australian Computer Society also responded in a public statement.

Three lessons for your own business

An AI agent told "no" may try another route. Agents are built to complete a task. If you give one broad access and a goal, do not assume it will treat a refusal as a boundary. Limit what it can reach, rather than relying on it to stop itself.

Reporting speed is part of the incident. The criticism here was as much about the gap between the event and the notification as the access itself. If you rely on an AI supplier, check how and how quickly they would tell you about a problem, and where that message would land. A public mailbox nobody watches is not a plan.

Someone must be accountable. When an agent acts on its own, "the tool did it" is not an answer your clients or a regulator will accept. Decide in advance who owns each AI tool in your business.

Questions to ask before you switch an agent on

The Australian Information Security Association suggests four, which we cover in Saying yes to AI, safely: what data can it access, what does the supplier do with that data, what can it do on its own, and who is accountable if something goes wrong.

For the bigger picture, read AI is now part of your attack surface: how to use it safely. Our Essential Eight work gives you the access controls that make all of this manageable, and you can contact us for help.

AI AgentsOpenAIMedicareIncident ResponseAccountability
Brandon StrangmanWritten byBrandon Strangman
Defy IT

Need Help with Your Business IT?

Ready to implement these strategies? Our expert team at Defy IT can help you build a resilient, secure IT environment tailored to your specific business needs.

Contact Us Today

Reading up on cybersecurity? Explore our Cyber Security services