AI Safety Warnings Intensify Amid Regulatory and Political Friction

Geoffrey Hinton, whose early research on neural networks underpins current large language models, has warned that frontier AI systems carry a material chance of escaping human control, estimating a 10 to 20 percent risk of human extinction in coming decades. He argues that self-regulation by companies is insufficient, comparing it to allowing tobacco or asbestos firms to set their own safety standards, and rejects the notion that uncertainty justifies inaction because a single failure would leave no opportunity for correction. Hinton also criticized President Trump’s use of the term “superintelligence” as evidence of limited technical grasp and described the president as a “corrupt, senile felon.” Nvidia chief Jensen Huang countered that Hinton’s probability figures lack scientific grounding and that alarmist claims hinder evidence-based progress. Separately, Anthropic introduced a narrow policy barring sustained, purposeless abusive language toward its Claude model, citing a possible moral status for advanced systems; the company has not released data on enforcement frequency or performance effects. These positions sit against a backdrop of rapid capability gains, documented safety incidents such as agent swarms on Hugging Face, and competing views on whether empathy toward models diverts attention from human harms linked to data-center expansion. Hinton noted that societies have historically required decades to absorb shifts that diminish human centrality, from Copernicus to Darwin, and that the emergence of non-human intelligences leaves even less time for adaptation.