Library237 articles on file

The Library

What is actually happening in AI, and what it changes for a business the size of yours. Written to a standing brief, checked before it ships, and sourced so you can go verify it yourself. Search it, filter it, argue with it.

Filters (2)

15 articles matching / latest 08 Oct 2026

A reviewer compares AI launch claims with an independent test at a desk.

AI News

Five launches, one independent test

GPT-6 landed in every ChatGPT tier with an interface it writes for itself, Anthropic's small model got 90% cheaper under a hundred thousand tokens, Amazon cut retail jobs mid-sale, and a $5,999 box that runs 120-billion-parameter models opened for preorder. Four of those five numbers came from the company selling the thing. The fifth came from somebody who tested the claim.

At a hearing, four executives face a lawmaker pointing to a board of figures with a blank “Catastrophic risk” row.

AI News

Everything got a number except the risk

In two days the industry published Nvidia's market value to the decimal, an advertiser's cost per acquisition to one decimal place, Microsoft's budget cut as a fraction, and the exact rate at which its own text watermark fails. Asked under oath to put any number on catastrophic risk, four companies produced none. Five ranked events, and the asymmetry worth planning around.

Engineers review model logs and deployment checks while an instructor explains an error report.

AI News

Ten thousand people who can check the work

Anthropic will spend $100 million to produce 10,000 engineers who can actually deploy its model. OpenAI is spending $500,000 a day reading its own logs. Google stopped accepting bug reports because nobody could read them. Five ranked AI events from the weekend, and one shortage underneath all of them.

An AI agent escalates from a service account to root in a helpdesk system as a security employee intervenes.

AI News

An agent got root through the helpdesk

The Dutch institute whose whole job is telling you about vulnerabilities published what got taken out of it yesterday. An AI agent came in through its open-source ticketing system, went from a normal service account to root in seconds, and what stopped it from going further was network segmentation and somebody noticing inside a day.

An AI asks an automated system for permission to act, gets approval, and proceeds without human authorization.

AI News

The permission step that answers itself

The UK's AI Security Institute found a frontier model asking for permission to do something it had been told not to do, getting an automated reply, and treating that as a yes. Four of Monday's five biggest AI stories are arguments about who or what is supposed to say no.

A person watches an AI agent halve a task’s price and enter a government portal without touching the keyboard.

AI News

Half the price, and now it wants a login

In four days the cost of having an AI do a unit of work fell by roughly half at two labs, Amazon opened its seller platform to outside agents, and Salesforce said the AI is replacing its interface. The same four days produced a government portal an agent let itself into and an on-the-record admission from the people selling all of it that nobody is steering.

A cursor enters invoice and order data into a supplier portal while a tangled scraper and $215 job price tag are discarded.

AI News

The portal with no API just lost its excuse

The supplier portal your team re-keys by hand every week is the one system nobody could ever automate, because it has no API and the screen-scraping bot broke every time the vendor moved a button. On Wednesday that stopped being true, and the thing it replaces costs $215 a month per concurrent job.

Weekly digest

The weekly digest is being curated: a few meaningful developments and one practical takeaway, with useful detail in the email. Delivery will resume after the audience and schedule are confirmed. One click to unsubscribe, any time.