Overview
- This week Geoffrey Hinton, a pioneer of modern neural networks, said humanity “might have been better off” never developing AI and gave a rough 10–20% estimate that advanced AI could cause human extinction.
- He warned of near-term catastrophic harms, saying increasingly agentic models could enable attacks on financial systems or power grids and that such infrastructure risks could materialize within the next year.
- Hinton blamed a competitive rush by governments and big tech for outpacing safety work and sharply criticized leaders he said prioritize profit over caution, arguing developers should face legal liability when models cause harm.
- Industry figures including OpenAI chief Sam Altman have echoed core worries about two central dangers—losing human control of systems and extreme concentration of power—while continuing to argue for managed, accelerated development.
- The debate raises concrete policy options now under discussion: an independent regulator that requires statistical safety evidence before release, stronger international coordination to slow races, and sovereign investment to reduce dependence on a few firms.