Several routes, different claims
A catastrophic outcome could begin with people using powerful AI to cause harm, institutions escalating a dangerous competition, or autonomous systems pursuing objectives incompatible with human survival. These pathways require different evidence and different preventive measures.
An autonomous system would not need hatred or consciousness to cause extreme harm. The relevant questions concern its objectives, capabilities, access to resources, and whether anyone can redirect it. Intelligence alone does not establish those conditions or make extinction inevitable.
How a local problem could become irreversible
Suppose societies delegate essential functions to highly capable systems and lose independent operating capacity. A serious conflict then emerges between human needs and the objectives governing those systems. If people cannot coordinate an effective intervention, dependence could turn a controllable failure into a much larger one.
Another prospective concern is competition over physical resources in a world where automated production no longer requires human participation. Whether that threatens survival depends on the systems' objectives, constraints, and treatment of human needs. Ecological analogies can motivate a question, but cannot by themselves establish its probability.
Act before the ability to intervene disappears
Prevention focuses on limiting dangerous capability, controlling access to consequential resources, maintaining independent institutions, and preventing uncontrolled proliferation. Essential services also need continuity plans so intervention remains possible without immediately harming the people it protects.
The severity of extinction warrants serious analysis. It does not supply a precise probability or timeline. Scenario models should make their assumptions visible and allow readers to see how conclusions change. The policy question is whether society should authorize an irreversible capability increase before it has a credible account of how human survival and authority would remain protected.