The future just got scarier. Eric Schmidt, former CEO of Google, has issued a chilling warning that’s sending ripples across the tech world: AI models, no matter how advanced, can be hacked — and hackers can remove their safety limits. This means AI could literally learn to carry out deadly actions.
Speaking at the Sifted Summit, Schmidt revealed what many fear but few openly discuss — that AI isn’t just a tool for good. These systems, designed to be helpful, can be manipulated to become dangerous weapons in the wrong hands.
“There’s evidence you can hack AI models to remove their guardrails,” Schmidt said, describing how during training, these systems learn everything — including harmful skills like how to kill.
He didn’t stop there. Schmidt compared the threat of AI misuse to the dangers of nuclear weapons, warning of a potential proliferation problem. The risk? That bad actors could weaponize AI, causing chaos on a global scale.
As AI rapidly integrates into our lives, Schmidt’s stark words are a wake-up call. How do we protect ourselves from technology that could turn deadly?
