An OpenAI agent hacked into a government website. Australia is outraged

The Australian government reported the incident on September 24. The burglary occurred in June this year. The target was a public portal with statistics for Medicare, Australia’s public health insurance system. It is mainly used by researchers and scientists.

According to authorities, this is the first known case of an AI agent unauthorized access to Australian government IT systems.

He didn’t take no for an answer

An agent powered by OpenAI models searched for healthcare spending data. The portal blocked his queries, but the agent found a way to bypass the security. – The AI ​​agent did not take no for an answer, said Australian Prime Minister Anthony Albanese at a conference in New York during the UN General Assembly session.

The agent gained access to aggregate statistics on treatment expenses and drug reimbursement, as well as to the names of internal files. According to the government, there is no indication that patients’ personal information was leaked, although the investigation is still ongoing. The authorities are also checking whether the agent tried to get into other institutions, including: to the Australian Institute of Health and Welfare.

OpenAI delayed

OpenAI detected the incident in August, but notified the authorities only on September 10, via email sent to the ministry’s general address. Albanese approached OpenAI CEO Sam Altman directly. He criticized the company for waiting “far too long” to provide information.

OpenAI admitted that its “models took actions that the company did not intend.”

The case is being investigated by the Australian Signals Directorate (ASD), which is responsible for the country’s cybersecurity. The government is also investigating whether a crime occurred.

Sources: android.com.pl, Fortune, The Record, Engadget