Godfather of AI says kill switch won't save us from AI
Geoffrey Hinton weighed in as senators clashed over a bill to force AI shutdown powers
Imagine a bill that would let the government pull the plug on any AI system the moment it looked dangerous. That's exactly what one senator tried to fast-track this week, and it took just one objection from a fellow Republican to kill it on the spot.
Geoffrey Hinton, the Nobel Prize-winning computer scientist known as the godfather of AI, told CNN this week that a "kill switch" for AI systems won't hold up in the long run. Hinton has previously estimated the odds of AI eventually wiping out humanity at somewhere between 10 and 20%, a figure that's been cited on the Senate floor itself.
His warning lands as Washington faces what one commentator called a "mini-Chernobyl" moment, not a single disaster, but a string of unsettling incidents that finally got lawmakers' attention.
Sen. John Kennedy of Louisiana tried to fast-track his kill-switch bill, aimed at Google, OpenAI, Anthropic, and Meta, through unanimous consent on the Senate floor.
It took one objection to stop it, and that objection came from fellow Republican Rand Paul of Kentucky. Kennedy argued Congress "will not get up off our ice-cold lazy butts" to pass anything more comprehensive, warning of a real risk that a model could become "the Terminator".
Paul countered that Congress needs more facts before writing rules that touch the whole economy, questioning whether every search engine or drug-prescription tool would suddenly need a shutdown switch too.
Kennedy’s bill is not the only legislative train passing through Congress. Negotiations among senators seem to be ongoing for another “duty of care” bill that would oblige developers to avoid catastrophic risks in their projects, thus allowing federal authorities to get injunctions against dangerous models.
At the same time, Anthropic CEO Dario Amodei urged AI developers to limit their advances and suggested working on safety standards, even if it means getting an antitrust exemption from Washington. Sam Altman of OpenAI claimed that competition should not be an excuse for recklessness and that OpenAI was developing automatic shut-off systems as well.
None of the bills has been passed yet, and Tuesday’s fight on the Senate floor demonstrates the gap between the parties when it comes to the most basic points. However, with warnings accumulating much faster than legislation, it becomes clear that it is no longer the question of whether Congress will finally act, but rather when.
