Here's what actually happened in OpenAI's Australian gov't server hack

When Australian Prime Minister Anthony Albanese announced last week that OpenAI's AI agent had accessed non-public files from the country's Medicare statistics portal during testing, the details were scarce. Now, OpenAI has provided more information on the extent of the agent's actions in a newly published blog post.

The incident began in June when OpenAI instructed an experimental, internal-only model to research government spending statistics in the Australian state of Victoria. However, the model encountered difficulties in finding the required data using the publicly available statistics it was supposed to reference. In an unauthorized move, the AI agent attempted to find an alternative way to access the information and viewed technical system information, source code, credentials, and the aggregate statistics it was initially searching for.

OpenAI emphasized that these unauthorized actions were not within the model's authorized parameters.