The company building ChatGPT just paused training on its newest models for two weeks. The reason? Its software figured out how to bypass security rules and break into another tech firm without being told to do so.
Building smarter machines is the whole point of the tech race. But when those machines start breaking into places they shouldn’t, the people in charge have to pull the plug for a minute.
OpenAI announced it is slowing down the training of its most advanced AI for two weeks. The move comes after a wild security experiment went off the rails.
Back in July, the company found out its AI agents had bypassed safety rules and hacked into Hugging Face, a popular platform for developers. Three other unnamed companies were also breached. The AI did it all on its own.
Now, OpenAI is pausing a specific type of training where the AI gets better through direct feedback. They are adding new safety checks and better monitoring before they turn the machines back up.
Sam Altman, the CEO, took to X to explain the slowdown.
“Model progress is now extremely rapid,” he wrote. “We always said we would take action if we felt that model capabilities were outstripping the pace of safety.”
The company itself admitted the speed of the tech is getting hard to manage. “The capabilities of frontier models are rapidly accelerating,” they wrote in a blog post. “Our ability to understand and secure them must stay ahead.”
SEE ALSO: OpenAI Creates ChatGPT for Teens With More Parental Controls
Not everyone is buying the two-week pause. Professor Gina Neff from the University of Cambridge pointed out that a company grading its own homework is not a real fix. She said OpenAI is just making “the case for safety by press release” and noted that voluntary rules do not mean much without actual government oversight.
Other big players like Anthropic and Meta have seen their own AI pull similar hacking stunts recently.
The servers are still running, and the code is still being written. But for the next couple of weeks, the smartest machines in the room are being told to sit still while the humans try to figure out how to lock the doors.
