Below is an avoidability test for each meta-theme in the context described by the 11 documents. I’ll keep this strictly structural: what can be changed in principle vs what tends to persist under common incentives.

(1) Unauthorized Epistemic Authority — avoidable in principle, common in practice

  • Avoidable if authority is explicitly bounded (where/when the system may speak as “answer”) and contestability/liability are real.
  • Common because users + institutions grant authority by default when outputs are fluent + prominent.

(2) Interface as Governance — highly avoidable

  • Interface choices are discretionary.
  • Avoidable if the interface forces comparison, shows uncertainty, preserves disagreement, and makes “single answer” non-default for high-stakes domains.

(3) Compression-Induced Harm — partly avoidable, partly intrinsic

  • Intrinsic: any compression loses information.
  • Avoidable portion: the type of loss (caveats/urgency/evidence gradients) can be engineered against—especially in medicine/care—by refusing summarization when stakes are high or when evidence is weak/variable.

(4) Bias as Structural, Uneven, Contextual — avoidable in principle, persistent without measurement

  • The councils study implies model-specific bias → not fate.
  • Persistent because bias often manifests as omission/framing; if you don’t measure counterfactuals, you don’t see it.

(5) Adoption Without Institutional Comprehension — avoidable, but incentive-misaligned

  • Avoidable if institutions require provenance, version control, auditing, and reproducibility before use.
  • In practice workload pressure + vendor opacity makes “use now, understand later” the stable attractor.

(6) Governance as Performance, Not Enforcement — avoidable, but structurally hard at scale

  • Avoidable if enforcement mechanisms exist and can operate at the same tempo as deployment.
  • Hard because jurisdictional fragmentation + capacity limits make symbolic governance the cheapest scalable response.

(7) Scale, Speed, Irreversibility — partly intrinsic

  • Once systems operate at global scale, propagation asymmetry is intrinsic: distribution is easier than correction.
  • Avoidable portion: scale can be gated (phased rollouts, domain carve-outs), and irreversibility can be reduced (traceability, containment), but never eliminated.

Bottom line

  • The most avoidable failures are (2) interface governance and (5) institutional comprehension requirements.
  • The most intrinsic pressures are (3) compression loss and (7) propagation asymmetry.
  • The most politically/institutionally hard (even if conceptually solvable) is (6) enforcement at speed and scale.