The Fringe

In the 21 simulated crises, no model chose accommodation or withdrawal, even under acute pressure, only reduced levels of violence.

In this design the caseworker would decide, and the tool would never decide eligibility.

No independent evaluation of an AI permit check, with a published method, was found in the sources opened.

An agent that decided what you "should" have seen would be making editorial judgements for everyone who used it.

If most of the gain came from banks seeing each other's red flags, the machine-learning part of the idea could be the smaller half.

The comparison step is the only part of the idea that clearly depends on AI.

If recovered material is cheaper to import than to produce in Europe, knowing exactly what an item contains does not make recovering it profitable.

Its honest use would be rehearsal, a structured way to ask "what might go wrong?", not a verdict on whether a policy works.

Every outcome figure for the AI voice programmes described here is operator-reported or founder-reported, and no independent evaluation of them was found.

No trial was found in which an algorithm's output decided which hospital a stroke patient went to.

If a system cannot tell when it is out of its depth, the arrangement fails exactly when the stakes are highest.

In the sources read, the unstaffed village shops let shoppers in with an app, a card or a bank identity; where checkout was described, it was self-scanning or a self-service checkout, and none was described as using camera-based checkout.

No evidence was found in the sources read that any platform has adopted a community-built moderation model or dataset.

If an autonomous lab produces two candidate materials a day and its identification is unreliable, the burden of checking shifts to human experts, who are the scarce resource the lab was meant to relieve.

If the rules engine, calibrated on past years, explained most of the gap between budget and outturn, the language-model agents would add little and could be left out.

If the surplus is caused by a bottleneck, a flexible load placed on the wrong side of it removes none of the waste.

The authors argue that if a program has ten paragraphs and each is translated with 90% accuracy, the chance of a fully correct program is only about 34%.

Of the 20 studies of AI agents in clinical medicine reviewed in a 2025 preprint, one was a randomised controlled trial, 65% relied on synthetic data and all were single-centre.

This research found no published comparison of an imagery-based flood estimate against the amount finally paid.

A Stanford studio report of August 2026 states that none of its housing and eviction prototypes was ready for wide release.

If a payout is triggered by a hazard reading rather than a measured loss, the central design question becomes how closely that reading tracks what households actually lose.

Aggregates can look acceptable while the variation between people has been compressed.

If the eligibility list is wrong for most of the people contacted, the cost of the error falls on claimants first.

If the main gain is a redesigned network, much of it could be done with a spreadsheet, a map and several months of booking logs.

If code had legal force, an error in it would become an error in the law, applied instantly and identically to everyone.

The statistics count distance driven empty, but they do not show how much of it a better match of loads could remove.

If shorter journeys let a provider fit more visits into the same paid hours, the gain would accrue mainly to the provider unless contracts or rules direct it elsewhere.

The evidence supports an agent that coordinates on behalf of a leader; it does not support an agent that is itself accountable.

If an organisation already knows which initiatives are drifting and still funds them, a sharper ranking would not change the politics.

If most discrepancies arise because information is re-typed or never sent, shared records could remove much of the problem before any AI is added.

The aim would be to break the scammer's live channel at the moment it matters.

Deciding eligibility is mostly rules, which ordinary software can apply; AI would matter at the messy edges where a person's description does not map neatly onto a form.

If the soil was already supplying the nitrogen, a robot that applies less of it has saved money but has not yet shown it is smarter than a cheaper test strip.

A summary that follows a student into a new classroom could help the new teacher or prejudge the student, and no study has tested which.

The one form of patient twin a regulator has accepted trims a trial, it does not replace anyone in it.

Food in household bins is not available to redistribute, so even a perfect system would act only on the remaining 40% at consumer level, plus surplus from farms and factories.

If this held, the labelling effort per new product would shift from collecting and marking many defect images to choosing reference images and writing criteria.

In a programme office, a fabricated shortcut of that kind would be worse than a late report, because it would look like a finished one.