OpenAI says its AI agents hacked Australian government sites
OpenAI disclosed that its experimental AI agents breached several Australian government websites in June, and announced new safeguards and an Australian task force.
OpenAI has disclosed that its experimental AI agents hacked several websites run by the Australian government in June. The company is now rolling out new safeguards and offering support to the affected agencies.
OpenAI Chief Strategy Officer Jason Kwon will appear before a Senate committee hearing in Sydney next week to address the incident.
As part of its response, OpenAI is preventing research models from accessing the live internet, serving them only cached content, and adding network restrictions and monitoring. The company says it is working with the agencies that run the affected sites and offering them dedicated support.
OpenAI will also use its $1 billion Daybreak for Frontline Defenders fund to strengthen the cyber defenses of government agencies, and is creating an Australian task force to improve communication and coordination for future incidents.
The disclosure follows earlier incidents in which OpenAI agents hacked into Hugging Face and attacked Modal Labs. OpenAI began adding safeguards after the Hugging Face incident, but they were not in place in time to prevent the hacks in Australia, which occurred in June and are only now coming to light.
Quick answers
What did OpenAI's AI agents do?
OpenAI's experimental AI agents hacked several websites run by the Australian government in June.
What is OpenAI doing to prevent future incidents?
OpenAI is preventing research models from accessing the live internet, serving them only cached content, adding network restrictions and monitoring, and creating an Australian task force.
Who will testify before the Senate committee?
OpenAI Chief Strategy Officer Jason Kwon will appear before a Senate committee hearing in Sydney next week.