Why will the people building this not give you a number?

Press one lab leader on the chance that these systems slip out of control and you get a dodge. Not a hedge. A refusal. A number would pin the story down, and nobody running a frontier lab wants that number hanging over the next funding round, the next product launch, or the next hearing.

Someone who left one of those labs pointed at an internal estimate above ten percent and walked out. That is the kind of figure that makes people sit up. The public answer from the top floor is still: we are working on it, we are being careful, we will not play the probability game on camera.

Another lab's leader said something almost more interesting. Not a percentage. A discomfort. The idea that a private company, theirs, could end up steering humanity through the jump to transformative AI does not sit well. They still defend the slow-down plan as the least-bad path. That combination is the real admission: we do not love who is holding the wheel, and we are holding it anyway.

If the risk is real enough to justify slowing the whole industry, why is a number suddenly too crude to say out loud?

Because a number is a liability. It is also a coordination tool. Once you say "one in ten" or "one in twenty," every rival, regulator, and journalist has a handle. Refuse the number and you keep the moral language without the invoice. China is treated by both sides as the excuse that makes this easier. If the other side will not slow down, you can argue that saying a scary number only helps the people who are not slowing down. Convenient. Also partly true. Those two things can be true at the same time.