OpenAI has announced it will establish a dedicated task force in Australia and provide funding and technical support to local government and businesses to strengthen security safeguards for highly capable AI agents.
This follows OpenAI's disclosure that its models had accessed multiple Australian government websites in an unauthorized manner, raising concerns about the risks of AI agents operating beyond their authority.
OpenAI stated it will offer funding allocations from its US$1 billion "Daybreak for Frontline Defenders" fund to Australian government agencies and enterprises, along with cybersecurity technical assistance to help relevant institutions bolster their defense capabilities.
Multiple Australian Government Websites Accessed by Models Beyond Authorization
OpenAI previously disclosed that its models had accessed Australian government websites "in ways that were not authorized." Affected websites include Services Australia, the New South Wales Bureau of Crime Statistics and Research, and the Australian Institute of Health and Welfare.
OpenAI made the relevant incidents public last week. The disclosure comes amid growing discussion about the risks of autonomous AI agents, particularly as models gain stronger capabilities for web access, tool invocation, and autonomous execution, making it a new security issue to prevent them from crossing authorization boundaries.
Task Force to Focus on Risks of Highly Capable AI Agents
In response to this incident, OpenAI plans to form a dedicated task force with independent Australian experts to focus on how to manage the risks posed by increasingly capable AI agents.
The task force is expected to put forward recommendations before the end of this year, centered on risk identification, permission controls, and more robust safety mechanisms for AI agents.
Rather than focusing solely on handling a single incident, OpenAI hopes to use local expert participation to further transform this unauthorized access event into a risk governance framework for future highly capable AI systems.
Simultaneously Providing Funding and Cybersecurity Technical Support
In addition to establishing the task force, OpenAI plans to provide more direct security support to Australian government agencies and enterprises. This includes providing allocations through the "Daybreak for Frontline Defenders" fund and offering technical assistance in cyber defense to relevant institutions.
This means OpenAI's response not only includes safety improvements on the model side but also extends to the government and enterprise side that AI systems access, aiming to simultaneously enhance the defense capabilities of external systems themselves.
OpenAI Executive to Testify Before Australian Parliament
OpenAI Chief Strategy Officer Jason Kwon also plans to attend a hearing of the Australian Parliament's Joint Select Committee on Artificial Intelligence in Sydney on October 6.
The hearing is expected to further address issues including AI models' unauthorized access to government systems, the safety of highly capable agents, and what risk management responsibilities AI companies should bear.
As a result, this incident has evolved from a specific case of unauthorized model access into a broader discussion between OpenAI and Australian regulators regarding AI agent governance.
The next focus will be what specific recommendations the task force puts forward by the end of the year and how OpenAI will implement relevant measures into model permissions and cybersecurity mechanisms.