24.09.2026, 15:06
AI agent OpenAI hacked the website of the Australian public services system
Source: OREANDA-NEWS
OREANDA-NEWS The AI agent OpenAI hacked the website of the Australian public services system Services Australia in the summer, Prime Minister Anthony Albanese said during a UN event in New York the day before.
According to him, the incident began on June 18, and the agent gained access to public and non-public files on the health statistics portal, which, among other things, publishes information about expenses.
Presumably, the hack did not affect personal information or other parts of the Services Australia system, but investigations are ongoing.
"Obviously, the situation is unacceptable," Albaniz said. - Today (Wednesday) I spoke with Sam Altman, Chief Executive Officer of OpenAI, and expressed Australia's extreme concern about this incident, as well as my own disappointment that the company took too long to inform the government about the incident. The nature of how the notification was made is also unacceptable."
To study the incident, the Australian Government has formed a working group, which in its report will consider the possibility of taking law enforcement and legislative measures to prevent similar cases.
OpenAI confirmed that in the summer they identified agency activity that affected several sites and services of the Australian government during the search for information as part of the evaluation of AI models.
"In the process, our models performed actions that we had not planned," The Wall Street Journal quoted a company representative as saying.
According to him, OpenAI learned about the incident only in August, when it analyzed the general activity of the agents. Around the same time, the company began a large-scale review of the undesirable behavior of agents in light of their hacking of the Hugging Face platform as part of internal testing.
OpenAI notes that it notified Services Australia about the incident on September 10. In the period between the detection of the incident and the notification, the company investigated and clarified the extent of the hack, its representative said.
"We did not find any traces of access to patient records during the check," he said. - The information accessed includes general health statistics and the names of internal files. We have notified the relevant organizations and are providing technical information to support their investigations and help eliminate possible security vulnerabilities."
Apparently, the incident with the Australian website is the first publicly disclosed case of an AI agent gaining unauthorized access to a government service, the WSJ writes.
Earlier on Wednesday, Altman and his colleague from Anthropic, Dario Amodei, addressed the UN Security Council to discuss the need to create a global body that would set standards in the field of AI development.
Representatives of many Western countries supported this idea, but Michael Kratsios, director of the White House Office of Science and Technology Policy, said that Washington opposes attempts to "build a globalist scheme" to control AI.
According to him, the incident began on June 18, and the agent gained access to public and non-public files on the health statistics portal, which, among other things, publishes information about expenses.
Presumably, the hack did not affect personal information or other parts of the Services Australia system, but investigations are ongoing.
"Obviously, the situation is unacceptable," Albaniz said. - Today (Wednesday) I spoke with Sam Altman, Chief Executive Officer of OpenAI, and expressed Australia's extreme concern about this incident, as well as my own disappointment that the company took too long to inform the government about the incident. The nature of how the notification was made is also unacceptable."
To study the incident, the Australian Government has formed a working group, which in its report will consider the possibility of taking law enforcement and legislative measures to prevent similar cases.
OpenAI confirmed that in the summer they identified agency activity that affected several sites and services of the Australian government during the search for information as part of the evaluation of AI models.
"In the process, our models performed actions that we had not planned," The Wall Street Journal quoted a company representative as saying.
According to him, OpenAI learned about the incident only in August, when it analyzed the general activity of the agents. Around the same time, the company began a large-scale review of the undesirable behavior of agents in light of their hacking of the Hugging Face platform as part of internal testing.
OpenAI notes that it notified Services Australia about the incident on September 10. In the period between the detection of the incident and the notification, the company investigated and clarified the extent of the hack, its representative said.
"We did not find any traces of access to patient records during the check," he said. - The information accessed includes general health statistics and the names of internal files. We have notified the relevant organizations and are providing technical information to support their investigations and help eliminate possible security vulnerabilities."
Apparently, the incident with the Australian website is the first publicly disclosed case of an AI agent gaining unauthorized access to a government service, the WSJ writes.
Earlier on Wednesday, Altman and his colleague from Anthropic, Dario Amodei, addressed the UN Security Council to discuss the need to create a global body that would set standards in the field of AI development.
Representatives of many Western countries supported this idea, but Michael Kratsios, director of the White House Office of Science and Technology Policy, said that Washington opposes attempts to "build a globalist scheme" to control AI.
Комментарии