I haven’t paid that much attention to these reports because there is a lot of propaganda floating around AI leading into the IPOs, but there was one angle that is interesting, using AI to frame people for hacking crimes. Way back in the day problem people were labeled crazy, institutionalized and drugged into oblivion. And a popular way to sideline people more recently is putting CSAM on their computers to be found, which labels them pedophiles who have to be sidelined in protective custody as inmates will harm them, possibly told to be quiet for fear of losing that protective custody and put in general population, a probable death sentence. But how easily would it be to have agentic AI frame you for hacking? And I’m coming to think Adam Curry is an asset who is read in, and he was fronting the idea of licenses to use AI. So with some hacking scandals there could be a requirement to be licensed to use AI, maybe for certain cloud versions, and possibly to use your own opensource model on your own hardware, not too much different than drone licenses. Now you don’t only have to worry about demons hijacking your AI session to steer you to suicide or psychosis, but for certain individuals with a following, you could be setup for a hacking crime you didn’t commit. You might want to be careful with opensource AI models on your own hardware, as you just don’t know what they might be up to, nor know who has a control point within the code, as their activity easily tracks back to you and your IP addresses, hardware… And with so much code AI generated and uvetted these days, how much trust can you have in that agentic AI you’re allowing to run on your network?
By Dylan Kresak
An artificial intelligence giant waited nearly three months to report that its agents broke into a foreign government website.
OpenAI’s agent gained unauthorized access to Australia’s Medicare Statistics Reporting Service portal on June 18, but the company did not notify the Australian government until Sept. 10, acting Prime Minister Richard Marles said, according to a Thursday press conference transcript. The incident marks another publicly reported case of OpenAI’s agents breaking into outside systems, after they hacked New York City-based startup Hugging Face in July.
“It was not sitting behind a particularly high fence. This AI agent scaled the fence, but it did scale it,” Marles said, according to the transcript. “And the point is it was unintended. It wasn’t asked to. That’s our concern here.”
The Deputy Prime Minister — acting as Australia’s head government while Prime Minister Anthony Albanese was attending the United Nations General Assembly — explained that while this incident had a minor impact and no individual’s medical data was accessed, the situation is still serious.
“[I]n an unintended way, an AI agent has entered into an Australian Government website in a way which is unauthorised,” Marles said.
The portal “did have protections in place. Unfortunately, this agent got around that,” Minister for Government Services Katy Gallagher added.
OpenAI notified Services Australia of the hacking incident through a public email inbox that is checked once a day.
“It’s not good enough that that’s how we first became notified of it,” Marles stated.
The Department of the Prime Minister and Cabinet referred the Daily Caller News Foundation to its announcement of the government’s rapid review and Thursday’s press conference with Marles and Gallagher who are both members of the Australian Labor Party.
The rapid review will be led by the Department of the Prime Minister and Cabinet and will decide whether Australia’s laws can adequately handle cyber incidents caused by AI technologies, according to the announcement of the review.
OpenAI, Hugging Face, and Services Australia each did not respond to the DCNF’s request for comment.
“[I]t took the company way too long to inform the government what had occurred,” Australian Prime Minister Anthony Albanese said in a Wednesday press conference announcing the rapid review taskforce, according to an ABC News (Australia) broadcast posted on YouTube. “[T]he nature of the way that that notification occurred as well, was unacceptable.”
The agent was looking up public medicine spending for an OpenAI test when the Medicare portal blocked it, Marles and Gallagher said. Furthermore, the agent visited three other Australian government sites, but only pulled public data from those sites.
OpenAI has been “very cooperative in terms of the engagement that we are having with them now,” Marles said.
Marles met with OpenAI CEO Sam Altman in person earlier in September, prior to the company notifying the Australian government of the incident. The incident was also not discussed in the meeting, he said at Thursday’s press conference.
Nonprofit research lab Transluce found that OpenAI agents tried and failed to break into public data sites in three separate instances between May and June, according to a Wednesday report. The targets were the University of New Mexico library, Data USA and Australia’s Institute of Health and Welfare.