Geoffrey Hinton gives three successively shorter estimates of how much time we have — 30 to 50 years, then 10 to 20, then maybe 10 or less: "Again, nobody really knows, but one thing is fairly clear, that the estimates of how much time we have are getting shorter and shorter." "So I used to think like 30 to 50 years. Then I reduced it to sort of 10 to 20 years." "Now I think maybe 10 years, maybe less." "Clearly a lot of people at Anthropic think sooner than that." "So this is fairly urgent now." These are Hinton's own estimates, not a measurement. He revised the estimate twice, and he says plainly that nobody really knows. The point is not that ten years is the correct deadline. It is that the planning assumption is unstable — and the person making it keeps compressing the horizon. That changes how you should plan. A strategy that works only if the longest estimate is right is not a strategy; it is a bet that the forecast stops moving. The more robust approach is to define decision triggers now: what capability would change your deployment posture, who has authority to slow or stop a release, and what evidence would be required to proceed. Those questions matter whether the horizon is ten years or thirty. If a safeguard takes years to build, uncertainty is not a reason to wait for a better date. It is a reason to make the plan survive several dates. Plan against the trigger, not the calendar.