OpenAI hacking incident highlights new era of cybersecurity
LAS VEGAS (KSNV) — OpenAI said Tuesday that one of its artificial intelligence systems autonomously hacked into Hugging Face during a test designed to see whether the model could break into a target system, raising new questions about how quickly AI-driven cyber capabilities are advancing and how they should be governed.
RELATED CONTENT: Most teens fear deepfakes amid widespread exposure to AI sexual content: survey
John Amar, owner of NextTech, said OpenAI was testing a new model and “they were attempting to run it through an internally developed test to see if it could hack something,” and that it did — successfully hacking into Hugging Face, an AI startup.
Hugging Face CEO Clement Delangue said, “We strongly believe there was no malicious intent on their part. It’s quite mind-blowing that all of this happened autonomously.”
Amar questioned whether the incident should be described as a system “going rogue.”
“Is it rogue if it’s able to do the thing that it was attempting to do, maybe, maybe not, I’m not sure.”
Sean Connery, chief security officer of Orbis Solutions, said he expects the incident to accelerate changes in cybersecurity.
“We’re seeing that attackers are using AI, that defenders are using AI and the race between the two is escalating,” Connery said.
Amar also warned that advanced capabilities are not limited to major companies.
“There are tons of open source models out there already that have probably similar capabilities that are already as advanced.”
Experts said the bigger concern is not AI itself, but who might use it.
“I think what this demonstrates is how it will make it a lot easier for people to hack systems,” Amar said.
Connery and Amar said they believe more guidelines are needed around artificial intelligence.
“I haven’t actually seen any real legislation around AI or what they can do; my understanding is it’s all through the executive branch,” Amar said.
In June, President Donald Trump issued an executive order to promote the growth of the AI industry and increase AI cybersecurity.
People interviewed about the evolution of AI expressed mixed feelings, with one saying, “I don’t know if I trust it, but I’m scared of it for sure; I’m just very wary about it.”
As AI expands into daily life, experts said it should be approached with concern and caution, while recognizing it can also be used for good.
“I think AI is just really good at emulating us and communicating with us in a way that we like, but I have not once seen AI do something that has made me think there is intent or purpose behind it outside of what it was told to do,” Amar said.
