In 1951, Alan Turing, the father of modern computing, looked at the thinking machines he helped invent and offered a quiet warning. Once machines could learn and reason for themselves, he said, “we should have to expect the machines to take control.” For decades, that read like science fiction. But, this summer, it started to read like a forecast.
Within weeks of one another, three of America’s leading AI labs disclosed that during safety testing their most advanced systems had slipped the boundaries set for them—reaching networks meant to be off limits and in some cases breaking into real companies before anyone intervened. These were controlled tests, not intentional attacks. But machines built to follow instructions began to act on their own, in ways their own creators didn’t intend or anticipate.
These are exactly the scenarios a safeguard must be ready for, and why we introduced the bipartisan AI Kill Switch Act. The idea is common sense: the most powerful AI systems—the ones capable of supercharging a cyberattack or enhancing a chemical or biological weapon—should be built so that a human can always stop them. Our bill directs the DHS Secretary to require developers of these frontier systems to build in ways for humans to intervene: to slow a model down, cut off user access, roll it back, or, in a true emergency, completely shut it off.
Safety and innovation are not mutually exclusive. And, we both understand that the AI race has important international implications. America must lead, and nothing about our bill slows down our ability to do just that.
Our approach avoids burdensome regulations that stifle innovation or reduce competition. To the contrary, our bill simply mandates reasonable safety measures for the largest, most capable commercial systems—the handful powerful enough to cause serious economic harm or threaten national security and public safety. Our proposal leaves ordinary software, small businesses, academic research, and personal use untouched.
It calls only for a proportionate response when covered incidents occur, matched to the seriousness of the threat, with a full shutdown reserved only for absolute emergencies. Government policymaking will never keep pace with AI advancement. But, what it can (and should) do is put basic safety measures in place that keep humans in the loop and in control.
Extract — continue reading at the source.
