According to the Seday-e Sama News Agency, citing ABC News, OpenAI announced on Friday that a second Australian government agency had been hacked by one of its AI agents. The incident was disclosed just one week after the company revealed its first intrusion into an Australian government system.

According to OpenAI, an AI agent gained access to the fire-records service of the New South Wales National Parks and Wildlife Service and extracted summary statistics about fires that were not publicly available.

OpenAI said the incident was identified during a broader investigation into activities by its models that were “misaligned with their intended purpose.” The company emphasized that, based on investigations conducted so far, there is no evidence that the AI agent accessed personal information.

New South Wales authorities said the incident likely occurred in June, but OpenAI notified the Australian government of the incident on Thursday. Several government agencies in the state have also launched investigations into the incident and are assessing its consequences.

This is the second disclosed case within a short period in which an OpenAI AI agent has infiltrated an Australian government agency.

In the previous incident, an AI agent accessed Australia’s Medicare statistics reporting system and gained access to some public and non-public files.

Australian Prime Minister Anthony Albanese had previously expressed serious concern over the incident and said that, in a conversation with OpenAI CEO Sam Altman, he had objected to the company’s handling of the notification. OpenAI also apologized for the incident.

The disclosure of these two incidents comes as OpenAI has faced a series of security concerns surrounding its autonomous AI agents in recent months. The company has also announced that it will cooperate with the Australian government to identify and address unwanted behavior by these agents.

— Tasnim