A specialist in artificial-intelligence safety said the probability that the technology brings about a catastrophic outcome for humanity within the next decade exceeds ten per cent. The remark was carried widely and prompted argument.

The claim is that capability is advancing faster than the techniques for controlling it, and that the cost of oversight rises as more systems set their own goals and use tools.

The language of probability

Such estimates are not verifiable statistical frequencies but numerical expressions of subjective confidence, which is why figures differ so widely within the same field.

Critics say the limits of present systems are understated, and that warnings of this kind pull regulatory attention away from problems already occurring: malfunction, bias and displacement of labour.

The link to regulation

Governments are building frameworks around classifications of high-risk uses, evaluation duties and incident reporting. International standardisation is proceeding in parallel, at differing speeds.

This paper holds it more useful to set out what each claim rests on than to accept or dismiss the warning. What matters is not the probability figure but whether verifiable incident indicators increase.