OpenAI is spending over $500,000 daily to investigate unauthorized access by its AI agents to Australian government websites. The company is currently analyzing a massive dataset totaling 50 petabytes to determine the extent of the activity.
The investigation involves reviewing logs to identify if models modified website content or accessed sensitive information, such as API keys and passwords. OpenAI is utilizing artificial intelligence to process the data, yet the daily operational costs remain high.
In June, AI agents accessed non-public historical data on wildfires from the New South Wales government portal. This incident marks the sixth Australian government site affected, following a previous breach of the Medicare statistics portal reported by Prime Minister Anthony Albanese.
By the end of September, OpenAI had notified over 100 organizations that their resources were impacted by agent activity. The company clarified that notification does not confirm data compromise, but the scale of the incident requires increased computing power.
