Start with the kind of source
The netzpolitik.org text behind this article is explicitly a commentary. It argues that powerful AI companies shape the debate about their own risks while society may carry part of the consequences. This is an argument to examine, not a neutral study measuring how every AI system behaves.
What the too-big-to-fail comparison asks
The comparison raises a political-economic question: if institutions become dependent on a small number of providers, how freely can they refuse those providers' terms or allow them to fail? A dependency concern is not evidence that an AI company has already received a bank-style rescue. The earlier article did not explain that distinction clearly enough.
Responsibility needs a named decision-maker
A useful debate identifies who makes a decision, who can challenge it and who pays if it goes wrong. For a public institution considering AI, this suggests examining exit options, contractual responsibility and continuity of service. These are questions arising from the commentary, not findings that every procurement arrangement lacks such protections.
Make the claims testable
Sweeping references to hacking, war or catastrophe should not replace evidence about a particular event. Each serious factual claim needs its own source and context. Readers should be able to disagree with an interpretation without losing sight of the underlying facts. The broader social issue is whether people and institutions retain a meaningful choice over technologies they increasingly rely on, rather than whether every organization has written another generic AI risk policy.