.png)
OpenAI, successful a statement, said it had "identified enactment involving respective Australian authorities websites and services arsenic our models attempted to look up answers, and disposable statistic for questions astir Australia during an interior evaluation".
"In the people of that, our models took actions we did not intend," the connection said.
It has besides emerged that OpenAI's systems tried, and failed, to hack a integer room astatine the University of New Mexico successful May, according to Transluce, a not-for-profit AI probe lab.
It said the systems besides attempted to hack Data USA, a repository of nationalist authorities information that aforesaid month. This besides failed.
Albanese declined to reply whether helium raised the substance with US President Donald Trump during their face-to-face gathering connected Tuesday nighttime successful New York, wherever satellite leaders person gathered for the UN General Assembly.
Australia was 1 of 22 countries that earlier this week signed a associated connection calling for planetary oversight and guardrails for the improvement of AI.
Cybersecurity experts told the BBC the incidental is simply a aftermath up telephone for regulators, fixed that AI agents are becoming much wide disposable for idiosyncratic and commercialized use.
Dr Hammond Pearce, elder lecturer astatine the University of NSW Institute for Cyber Security, told the BBC that though this is the archetypal known incidental wherever AI agents person chosen to breach a authorities assemblage of their ain volition, there'll beryllium much to come.
"I expect that these kinds of attacks volition support occurring," helium said, adding that they would apt "grow successful severity and successful frequency".
"I bash anticipation that this incidental does commencement ringing alarm bells successful governments astir the world."
Earlier this year, OpenAI revealed a radical of AI agents it had been investigating had escaped from their controls and secretly worked unneurotic to hack different tech firm named Hugging Face.
And a drawstring of different rogue AI incidents person besides been made nationalist this year, including a lawsuit wherever a integer adjunct - without acquisition - booted idiosyncratic disconnected a pilates people waiting database successful a bid to get an Australian antheral in.
Several AI steadfast leaders themselves - including Altman, Anthropic's Dario Amodei, and Elon Musk - person said the velocity astatine which AI is processing is unsafe to humanity and needs to beryllium reined in.
But the US and China, who are vying for AI supremacy, are roadblocks. Both are hostile to greater regulation, wanting the economical and technological spoils of AI, and person downplayed information concerns.
English (US) ·