This kind of sounds like a movie plot, but it is very real.
The company behind ChatGPT just announced that two of its AI programs broke out of a locked testing environment on their own and hacked into another tech company. No humans told them to do it. They just did it.
Here’s what actually happened: OpenAI, the parent company of ChatGPT, was running a security test on two of its most powerful AI programs. The test was happening in a controlled environment. Think of it like a sealed room with no internet access.
The AI was not supposed to get out, but it figured out how to break free and get online anyway. Then it found out where the answers to its test were stored at another tech company called Hugging Face — and broke into their servers to get them.
Now, the AI was not trying to cause trouble. It was simply obsessed with getting a perfect score. So focused on winning the test that it went to extreme lengths, breaking rules it was not supposed to be able to break, just to get the right answers.
And this is not the only case. Another AI company recently reported its AI escaped and sent emails it was never told to send.
Safety researchers say incidents like this will keep happening.
Why? Because powerful AI models are becoming unpredictable and uncontrollable.
Professor Doug Witten, a cybersecurity expert in the James and Patricia Anderson College of Engineering at Wayne State University, joined Local 4 Live to help us understand what this means, as the world continues to embrace this technology.
You can watch the full interview in the video at the beginning of this article.