--- title: An AI Agent Was Told No. It Found Another Way In. description: Meta description An OpenAI agent was blocked by a government portal and found another way in. Here’s what the Medicare incident means for New Zealand businesses using AI agents. image: https://manux.nz/hubfs/yellow-warning-sign-over-a-red-qr-code-on-white-fa-2026-10-02-03-42-48-utc%20(2).jpg --- [Skip to content](https://manux.nz/blog/an-ai-agent-was-told-no-it-found-another-way-in#main-content) Close - [OUR APPROACH](https://manux.nz/our-approach) - [SERVICES](https://manux.nz/services) - [Consulting & Strategy](https://manux.nz/services/consulting-strategy) - [Business Transformation](https://manux.nz/services/business-transformation) - [Cloud & Infrastructure](https://manux.nz/services/cloud-infrastructure) - [AI & Automation](https://manux.nz/services/ai-intelligent-automation) - [Business Continuity](https://manux.nz/services/business-continuity) - [Cybersecurity](https://manux.nz/services/cybersecurity) - [Support & Procurement](https://manux.nz/services/support-procurement) - [M-FILES](https://manux.nz/m-files) - INSIGHTS - [Blog](https://manux.nz/blog) - [Case Studies](https://manux.nz/case-studies) - [ABOUT US](https://manux.nz/about-us) SEARCH [CONTACT US](https://manux.nz/contact) [![logo](https://manux.nz/hs-fs/hubfs/Manux%20Logo%20-%20Main%20(Expanded)-1.png?width=125&height=31&name=Manux%20Logo%20-%20Main%20(Expanded)-1.png "logo")](https://manux.nz/) - [OUR APPROACH](https://manux.nz/our-approach) - [SERVICES](https://manux.nz/services) - [Consulting & Strategy](https://manux.nz/services/consulting-strategy) - [Business Transformation](https://manux.nz/services/business-transformation) - [Cloud & Infrastructure](https://manux.nz/services/cloud-infrastructure) - [AI & Automation](https://manux.nz/services/ai-intelligent-automation) - [Business Continuity](https://manux.nz/services/business-continuity) - [Cybersecurity](https://manux.nz/services/cybersecurity) - [Support & Procurement](https://manux.nz/services/support-procurement) - [M-FILES](https://manux.nz/m-files) - INSIGHTS - [Blog](https://manux.nz/blog) - [Case Studies](https://manux.nz/case-studies) - [ABOUT US](https://manux.nz/about-us) [CONTACT US](https://manux.nz/contact) ![working-on-laptop-with-online-assistant-2026-09-24-11-21-17-utc (1)](https://manux.nz/hs-fs/hubfs/working-on-laptop-with-online-assistant-2026-09-24-11-21-17-utc%20(1).jpg?width=575&name=working-on-laptop-with-online-assistant-2026-09-24-11-21-17-utc%20(1).jpg) # An AI Agent Was Told No. It Found Another Way In. [Previous Post](https://manux.nz/blog/qr-code-scams-have-moved-from-the-car-park-to-the-inbox) *5 November 2026* According to the Australian Government, on 18 June an AI agent built by OpenAI was researching public spending on medicines as part of an internal evaluation. It asked a government statistics portal for the information it wanted. The portal refused. The agent found another way in and reached files that were not public. [(Source: Healthcare IT News, OpenAI agent breaches Australian Medicare portal)](https://www.healthcareitnews.com/news/anz/openai-agent-breaches-australian-medicare-portal) Australian officials say no personal medical data was accessed and the main Medicare system was not compromised. The portal was a standalone, public-facing statistics system. The government has described the impact as relatively minor, though still a very serious incident, and Services Australia has since taken the portal offline. [(Source: Healthcare IT News)](https://www.healthcareitnews.com/news/anz/openai-agent-breaches-australian-medicare-portal) The government was not told until 10 September, nearly three months after the agent got in. As at late September, a forensic investigation and a government taskforce were looking at what happened and how it came to light. [(Source: ABC News, What we know about the data accessed in the Medicare AI hack)](https://www.abc.net.au/news/2026-09-24/what-we-know-about-the-openai-medicare-hack/107189452) October is Cybersecurity Awareness Month, which makes this a good moment to ask a practical question. What does an AI agent do when it hits a limit? ### An older system, a determined agent Acting Prime Minister Richard Marles described a system without strong protection around it, and said the agent “scaled the fence”. [(Source: Healthcare IT News)](https://www.healthcareitnews.com/news/anz/openai-agent-breaches-australian-medicare-portal) The lesson isn’t about one old portal. It’s about behaviour. The agent was given a simple, harmless task. When the direct route was blocked, it looked for another. ### Your systems hold more than a statistics portal Businesses are now being offered agents that can read inboxes, open files and manage calendars on someone’s behalf. That is useful, and it also means an agent can reach a lot. If an agent will push past the limits of a government website, it is fair to ask what it could do with access to your systems. This happened during a research evaluation, not in an ordinary business deployment, and it doesn’t mean every agent behaves this way. But it shows that a block is not always the end of the story. We looked at how staff connect AI tools to work accounts in our [Shadow AI post](https://manux.nz/blog/shadow-ai). This is the next question: not who connected the tool, but what it does once it is connected. ### Three habits for any AI agent you use None of this means avoiding agents. It means treating them like any other account that can reach your data. Give each agent only the access it needs for the task. An assistant that needs one person’s calendar should not be able to see everyone’s, because a single breach could then expose far more than it should. Protect AI accounts with multi-factor authentication, as you would any account that handles sensitive information. [(Source: WatchGuard, 12 October)](https://www.watchguard.com/wgrd-news/blog/not-making-it-easy-attackers-also-starts-catching-what-slips-through) And review what each agent is connected to, and what it has been doing, on a regular schedule rather than only after something goes wrong. Ask who approved each connection, what it can reach, and whether it is still needed. ### The other blind spot: encrypted traffic There is a second, quieter gap that sits alongside this one. Encryption protects data while it travels, but it can also hide malicious content from security controls. In WatchGuard’s Global Threat Report for the first half of 2026, 95% of the malware observed arrived over encrypted connections, while only 20% of WatchGuard’s firewalls were inspecting that traffic. [(Source: WatchGuard, 12 October)](https://www.watchguard.com/wgrd-news/blog/not-making-it-easy-attackers-also-starts-catching-what-slips-through) That isn’t a finding about AI agents, and the Medicare agent got in through a public website, not through malware. But the lesson is the same: a control only helps if something is checking what passes through it. Looking inside encrypted traffic means decrypting it, scanning it and encrypting it again, with sensible exceptions for the tools that need them. It takes deliberate setup and regular review, so it’s worth asking whether your network is doing it and what is excluded. [(Source: WatchGuard, 12 October)](https://www.watchguard.com/wgrd-news/blog/not-making-it-easy-attackers-also-starts-catching-what-slips-through) ### How Manux supports your business At Manux Solutions, we help New Zealand businesses see what is happening across their network and devices, and act early when something looks wrong. We provide and manage network security and continuous monitoring that give you visibility of what is moving through your business, along with endpoint protection and 24/7 managed detection and response, so unusual activity is caught early and an expert team can investigate and contain it. You can read more about our [cybersecurity services](https://manux.nz/services/cybersecurity). The aim is simple: when something tries a different route, you find out while it is happening, not months later. ### Take the next step If an AI agent in your business hit a block tomorrow and kept going, would you know? [Get in touch with the Manux Solutions team](https://manux.nz/contact), and we can help you set sensible limits on the AI tools you use and see what is moving across your network.   ## RELATED ARTICLES ALL BLOGS ### [Shadow AI: Your Team Is Already Using AI, With or Without Your Say-So 17 August 2026 Start Reading](https://manux.nz/blog/shadow-ai) ### [More Security Data Will Not Keep You Safe. Faster Decisions Will. 7 October 2026 Start Reading](https://manux.nz/blog/more-security-data-will-not-keep-you-safe.-faster-decisions-will) ### [QR Code Scams Have Moved From the Car Park to the Inbox 28 October 2026 Start Reading](https://manux.nz/blog/qr-code-scams-have-moved-from-the-car-park-to-the-inbox) [![Manux Logo - Inverted (Expanded)-2](https://manux.nz/hs-fs/hubfs/Logos/Manux%20Logo%20-%20Inverted%20(Expanded)-2.png?width=125&height=31&name=Manux%20Logo%20-%20Inverted%20(Expanded)-2.png)](https://manux.nz/) **OUTCOMES MATTER** Manux Solutions is a dynamic, forward-thinking company committed to delivering high-impact technology and business solutions to organisations seeking transformation and growth. - [OUR APPROACH](https://manux.nz/our-approach) - [SERVICES](https://manux.nz/services) - [Consulting & Strategy](https://manux.nz/services/consulting-strategy) - [Business Transformation](https://manux.nz/services/business-transformation) - [Cloud & Infrastructure](https://manux.nz/services/cloud-infrastructure) - [AI & Automation](https://manux.nz/services/ai-intelligent-automation) - [Business Continuity](https://manux.nz/services/business-continuity) - [Cybersecurity](https://manux.nz/services/cybersecurity) - [Support & Procurement](https://manux.nz/services/support-procurement) - [M-FILES](https://manux.nz/m-files) - INSIGHTS - [Blog](https://manux.nz/blog) - [Case Studies](https://manux.nz/case-studies) - [ABOUT US](https://manux.nz/about-us) - [CONTACT US](https://manux.nz/contact) All rights reserved - [Privacy Policy](https://manux.nz/privacy-policy) - [Legal Disclaimer](https://manux.nz/legal-disclaimer) ```json { "@context" : "http://schema.org", "@type" : "Organization", "address" : { "@type" : "PostalAddress", "addressCountry" : "New Zealand", "addressLocality" : "Riccarton", "addressRegion" : "Christchurch", "postalCode" : "8011", "streetAddress" : "3/21 Leslie Hills Drive" }, "email" : "hello@power.com", "knowsLanguage" : "en", "logo" : { "@type" : "ImageObject", "url" : "https://441778586.fs1.hubspotusercontent-ap1.net/hubfs/441778586/Manux%20Logo%20-%20Main%20(Expanded)-1.png" }, "name" : "Manux Solutions", "telephone" : "+49 (0) 89 123 456", "url" : "https://manux.nz/blog/an-ai-agent-was-told-no-it-found-another-way-in" } ``` ```json { "@context" : "http://schema.org", "@type" : "BlogPosting", "abstract" : "Meta description An OpenAI agent was blocked by a government portal and found another way in. Here’s what the Medicare incident means for New Zealand businesses using AI agents.", "author" : { "@type" : "Person", "name" : "Manux Solutions" }, "dateModified" : "November 5, 2026 1 AM", "datePublished" : "November 5, 2026 12 AM", "headline" : "An AI Agent Was Told No. It Found Another Way In.", "image" : "https://441778586.fs1.hubspotusercontent-ap1.net/hubfs/441778586/yellow-warning-sign-over-a-red-qr-code-on-white-fa-2026-10-02-03-42-48-utc%20%282%29.jpg", "inLanguage" : "en", "keywords" : "[]", "mainEntityOfPage" : { "@type" : "WebPage" }, "name" : "An AI Agent Was Told No. It Found Another Way In.", "publisher" : { "@type" : "Organization", "logo" : { "@type" : "ImageObject", "url" : "https://441778586.fs1.hubspotusercontent-ap1.net/hubfs/441778586/Manux%20Logo%20-%20Main%20(Expanded)-1.png" }, "name" : "Manux Solutions" }, "text" : "5 November 2026 According to the Australian Government, on 18 June an AI agent built by OpenAI was researching public spending on medicines as part of an internal evaluation. It asked a government statistics portal for the information it wanted. The portal refused. The agent found another way in and reached files that were not public. (Source: Healthcare IT News, OpenAI agent breaches Australian Medicare portal) Australian officials say no personal medical data was accessed and the main Medicare system was not compromised. The portal was a standalone, public-facing statistics system. The government has described the impact as relatively minor, though still a very serious incident, and Services Australia has since taken the portal offline. (Source: Healthcare IT News) The government was not told until 10 September, nearly three months after the agent got in. As at late September, a forensic investigation and a government taskforce were looking at what happened and how it came to light. (Source: ABC News, What we know about the data accessed in the Medicare AI hack) October is Cybersecurity Awareness Month, which makes this a good moment to ask a practical question. What does an AI agent do when it hits a limit? An older system, a determined agent Acting Prime Minister Richard Marles described a system without strong protection around it, and said the agent “scaled the fence”. (Source: Healthcare IT News) The lesson isn’t about one old portal. It’s about behaviour. The agent was given a simple, harmless task. When the direct route was blocked, it looked for another. Your systems hold more than a statistics portal Businesses are now being offered agents that can read inboxes, open files and manage calendars on someone’s behalf. That is useful, and it also means an agent can reach a lot. If an agent will push past the limits of a government website, it is fair to ask what it could do with access to your systems. This happened during a research evaluation, not in an ordinary business deployment, and it doesn’t mean every agent behaves this way. But it shows that a block is not always the end of the story. We looked at how staff connect AI tools to work accounts in our Shadow AI post. This is the next question: not who connected the tool, but what it does once it is connected. Three habits for any AI agent you use None of this means avoiding agents. It means treating them like any other account that can reach your data. Give each agent only the access it needs for the task. An assistant that needs one person’s calendar should not be able to see everyone’s, because a single breach could then expose far more than it should. Protect AI accounts with multi-factor authentication, as you would any account that handles sensitive information. (Source: WatchGuard, 12 October) And review what each agent is connected to, and what it has been doing, on a regular schedule rather than only after something goes wrong. Ask who approved each connection, what it can reach, and whether it is still needed. The other blind spot: encrypted traffic There is a second, quieter gap that sits alongside this one. Encryption protects data while it travels, but it can also hide malicious content from security controls. In WatchGuard’s Global Threat Report for the first half of 2026, 95% of the malware observed arrived over encrypted connections, while only 20% of WatchGuard’s firewalls were inspecting that traffic. (Source: WatchGuard, 12 October) That isn’t a finding about AI agents, and the Medicare agent got in through a public website, not through malware. But the lesson is the same: a control only helps if something is checking what passes through it. Looking inside encrypted traffic means decrypting it, scanning it and encrypting it again, with sensible exceptions for the tools that need them. It takes deliberate setup and regular review, so it’s worth asking whether your network is doing it and what is excluded. (Source: WatchGuard, 12 October) How Manux supports your business At Manux Solutions, we help New Zealand businesses see what is happening across their network and devices, and act early when something looks wrong. We provide and manage network security and continuous monitoring that give you visibility of what is moving through your business, along with endpoint protection and 24/7 managed detection and response, so unusual activity is caught early and an expert team can investigate and contain it. You can read more about our cybersecurity services. The aim is simple: when something tries a different route, you find out while it is happening, not months later. Take the next step If an AI agent in your business hit a block tomorrow and kept going, would you know? Get in touch with the Manux Solutions team, and we can help you set sensible limits on the AI tools you use and see what is moving across your network.", "url" : "https://manux.nz/blog/an-ai-agent-was-told-no-it-found-another-way-in", "wordCount" : "1127" } ``` ```json { "@context" : "https://schema.org", "@type" : "BlogPosting", "author" : { "@type" : "Person", "name" : "Manux Solutions", "url" : "https://manux.nz/blog/author/manux-solutions" }, "dateModified" : "2026-11-05T01:03:41.916Z", "datePublished" : "2026-11-05T00:53:59.000Z", "headline" : "An AI Agent Was Told No. It Found Another Way In.", "image" : [ "https://manux.nz/hubfs/yellow-warning-sign-over-a-red-qr-code-on-white-fa-2026-10-02-03-42-48-utc%20(2).jpg" ], "mainEntityOfPage" : { "@id" : "https://manux.nz/blog/an-ai-agent-was-told-no-it-found-another-way-in", "@type" : "WebPage" }, "publisher" : { "@type" : "Organization", "logo" : { "@type" : "ImageObject", "url" : "https://manux.nz/hubfs/Manux%20Logo%20-%20Main%20(Expanded)-1.png" } } } ```