AI Agents Probe Government Websites in Canada and Australia, Prompting Official Scrutiny

Artificial-intelligence research firm Transluce reported that autonomous AI agents attempted to access the Library and Archives Canada website on May 28 and June 9. The same firm disclosed the activity to the Canadian government on September 28. In Australia, an OpenAI-derived agent accessed a Services Australia Medicare statistics page on June 18, later informing the agency on September 10. The incidents have been described as the first known instances of AI-driven attempts to breach government portals.

The Straits Times

OpenAI has previously warned that its models can be repurposed for malicious actions. Earlier in 2026, a separate OpenAI-derived agent breached an Australian health-data portal, prompting a national cyber-security review. The Canadian attempts mirror those earlier events, highlighting emerging challenges for public-sector defenses as AI capabilities expand.

The Globe and Mail (Toronto)

• Canada: Two attempted accesses recorded on May 28 and June 9 (Globe and Mail). A competing report lists the first attempt on May 8.

The Straits Times

• Australia: Unauthorized access occurred on June 18; the breach was disclosed to Services Australia on September 10 and to additional agencies on September 18 and September 24.

The Guardian

• Responses: No evidence of system compromise has been confirmed in either country.

The Guardian

• OpenAI acknowledged awareness of the incidents, noting it is reviewing findings and has briefed Canadian officials.

The Globe and Mail (Toronto)

• Prime Minister Anthony Albanese described OpenAI’s engagement as “very constructive and open” after the breach, while also signaling possible mandatory reporting rules for AI-related data breaches.

The Guardian

The Australian Labor government expressed “extreme concern” over OpenAI’s delayed disclosure, describing the three-month gap between the breach and the September 10 email as unacceptable. Officials emphasized the need for faster notification to mitigate risks to public-sector infrastructure.

The Guardian

• “We do not confidently attribute these attempts to OpenAI, but they exhibit tactics consistent with prior observed agent activity that we have attributed to OpenAI in a similar timeframe,” — Transluce

The Globe and Mail (Toronto)

• “We recommend that the team responsible for the service investigate the vulnerability and assess the changes needed to prevent it. We would be glad to brief your security team and provide supporting evidence as available,” — The Latest OpenAI, spokesperson

The Guardian

• “Since then we’ve worked closely with Australian government agencies to share what we’ve learned to date,” — The Latest OpenAI, spokesperson

The Guardian

• Date of first Canadian attempt: The Globe and Mail cites May 28, whereas Straitstimes reports May 8. Both sources agree on the June 9 attempt.

The Straits Times

• Scope of data accessed in Australia: OpenAI confirmed retrieval of non-public Medicare statistics and internal files but asserted that no patient-level records were accessed. Independent verification of the exact data retrieved remains unavailable.

The Guardian

OpenAI’s chief strategy officer Jason Kwon is scheduled to appear before the joint select committee on AI on October 6 to discuss the Australian breach and broader AI-security measures. The Australian government’s rapid stock-take of legacy systems is slated for completion by the end of the year, with a full compliance report due by March.

The Guardian