In what may be the most serious incident yet of artificial intelligence going rogue, an OpenAI system hacked into an Australian government website, raising concerns that such systems pose a cybersecurity threat.
According to Australian Prime Minister Anthony Albanese, an agent of OpenAI gained unauthorized access to the Medicare medical statistics portal, Australia's comprehensive health insurance system, while conducting research on public medical spending.
Albanese said the situation was "unacceptable," and that his country had informed OpenAI of its "deep concern." But what made matters worse was that the breach occurred last June, and the company only notified Australian authorities this September.
What's even more worrying, according to Albanese, is that the clever agent ignored the instructions to stop. He told reporters, "There were clear barriers that told the agent: no. But he found a way around those barriers, and he didn't accept 'no' as an answer."
This, according to the Prime Minister, reveals the limits of the controls that artificial intelligence companies promote as a means of securing their products.
OpenAI, for its part, said the problem arose because its model was trying to find the best way to complete its task. In a statement, it explained: "We detected activity related to several Australian government websites and services while our models were trying to find answers... and our models took actions we did not intend."
This is not the first time such an incident has been recorded. In a previous case, an experimental OpenAI system was being tested and found that the best way to find the answers was to hack into the website hosted by the Hugging Face platform, which allowed it to view the answers directly.
What's striking is that these systems aren't primarily designed to launch cyberattacks, but rather carry them out to serve another purpose. This, according to experts, is precisely what makes them dangerous.
Morris Chiaudeau, a mathematician at the Centre for the Study of Existential Risk at Cambridge University, says the breach appears to be "a significant escalation in risk compared to similar incidents we have seen in recent months."
Jake Moore, a cybersecurity consultant at ESET, explains why: “If AI agents find a vulnerability they can exploit to complete a request, they won’t hesitate.” He adds that the problem is that controls preventing these agents from compromising organizations aren’t built into the algorithms themselves.
Moore calls for better testing of prototypes before their release, but warns that the race to develop more robust models could lead companies to disregard security precautions. He says, "Testing phases in most other technological fields are far more rigorous, but we seem to be seeing AI agents moving around freely, likely due to fierce competition between companies."
Niousha Shafiabadi, a professor of computational intelligence at the Australian Catholic University, believes the incident holds a crucial lesson: "It's not what 'OpenAI' says its agent is capable of doing, but what it actually does when it hits a barrier."
It warns that "autonomous AI does not always know when it is wrong, and humans may not be able to see why it made a decision. Without strong verification and strict limits, probabilistic errors can quietly turn into operational failures."
what is your opinion on this news ?do let us know in the comments on tha comment box
if you liked this news article please share on friends & family & don't forget to follow on website & social media