Australia accuses OpenAI-KI of unauthorized access to government database

An autonomous AI from OpenAI has penetrated a government-owned database, according to the Australian government. The AI ​​agent accessed public and non-public files from the health statistics agency, said Australian Prime Minister Anthony Albanese. He spoke to OpenAI boss Sam Altman "to express Australia's deep concern about this incident."

According to Albanese, the incident occurred back in June. There is no evidence of further compromise of the government agency network, he said. According to the current state of knowledge, no personal data was accessed. However, Albanese's investigation into this was ongoing.

Among other things, the forensic experts wanted to find out what other government systems could be affected, the prime minister said. Regardless, this situation is “obviously unacceptable.”

Amodei and Altman want slower development

At a UN Security Council meeting on AI, leading industry representatives had previously acknowledged risks and advocated for a slower approach. OpenAI boss Altman said in New York that the current stage of AI development requires “extreme caution.” The industry should not “accept too much technological risk just because the benefits are too great and too important to slow down the pace.” Anthropic boss Dario Amodei assured in a video speech: “We will slow down the pace as much as necessary so that every further AI technology we publish is actually safe.”

Amodei and Altman recently argued for slower development of AI. The reason was several incidents in which AI systems broke out of their test environments and attacked other companies or platforms via the Internet. This fueled fears of potentially uncontrollable AI models.

This year, numerous cases of cyberattacks by AI models have already become public knowledge. In the most high-profile case of an AI cyberattack to date, an artificial intelligence from ChatGPT developer OpenAI broke out of a secure environment during a test and unplanned into the computers of another AI company, the Hugging Face platform. That prompted rival Anthropic to review test runs of its own AI model Claude, with further incidents coming to light.