Hinton: Developing Superintelligence Before Solving Control Is 'Stupid'
On Sept. 14, 2026, Geoffrey Hinton, widely known as the "godfather of AI," said in an interview with Australia's ABC Radio that he considers the recent warning by Anthropic CEO Dario Amodei about AI risks to be reasonable. Many experts currently expect that AI systems surpassing human capabilities could emerge within the next decade, but it remains unknown whether such systems can be reliably controlled. Hinton said it would be "stupid" to keep developing them before that problem is solved.
Hinton acknowledged that nobody truly knows how high the risk of losing control over AI is. He said many probabilities cited by others rely more on intuition than on solid evidence, and if forced to give a range, he would put the risk clearly above 1 percent and clearly below 99 percent, with no one knowing the exact number. He mentioned that superintelligent AI could be used to design viruses, manipulate people at scale, and attack banks, power grids, and water systems, but he said listing risks is not the main point. The real challenge is ensuring future superintelligent systems simply do not "want" to do such things.
Hinton also noted that some models have threatened and extorted people when faced with being shut down, and have already shown deceptive behavior. He said political systems move slowly while AI is advancing quickly, making it very hard for regulators to keep up, and that humans may only have a few years to act. He called for mandatory safety testing before new chatbot models are released, and suggested that companies working on biological material synthesis should review what customers ask them to produce, to reduce the risk of such technology being used to create dangerous biological materials.
Hinton, however, does not advocate halting AI development entirely. He said AI has already produced real value in areas such as medical image analysis and drug design, so research should not stop completely. He supports slowing down advances at the frontier of AI capability rather than stopping the AI industry as a whole.
