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.
11 articles matching / latest 29 Sept 2026
AI News
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.
AI News
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.
AI News
Anthropic shipped 43 ready-to-run small business workflows yesterday, including one that answers an after-hours web form, qualifies it, books it, and logs it. The connector counts underneath each one are the real story, and so is the input nobody mentions.
AI News
The hundred dollar subscription you build your own tools on holds about a sixth less building per week starting today, at the same price. Anthropic announced it as a 25 percent increase, and it is, against a baseline nobody has seen since May.
AI News
Anthropic shipped a way to keep an AI agent's instructions in a text file, with an approval step before a change takes effect and a diff after it does. The thing it replaces is the automation nobody can explain and nobody dares turn off.
AI News
Anthropic cut the price of re-reading context on Fable 5.1 by 75 percent, down to a fortieth of what a fresh read costs. The saving lands in proportion to how tightly your work is packed in time, and the penalty for breaking a cache quietly quadrupled on the same day.
AI News
Claude now carries one memory across chat and Cowork, and exposes it as short files you can open and correct. The catch is that the file belongs to a person, not to a company, and there is no way to publish the canonical one.
AI News
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.
AI News
Anthropic's most capable generally available model cannot be run with zero data retention at all, by policy, on any platform. On Wednesday OpenAI announced it is building the machinery specifically so it never has to ask you for that trade. Which vendor holds your client files, and for how long, is now something you can look up before you build.
AI News
Every Claude model launched since August 2 weaves an invisible watermark into the text it generates, worldwide, and it travels when the text is copied and pasted. The detector that reads it has not shipped yet, which means for now the mark is a liability you carry and not a check you can run.
AI News
On August 14 Claude Code stops asking permission before every command on Pro, Max, and Team plans. Anthropic published the study behind the decision, and the finding is not that the machine got smarter. It is that in a controlled test, the humans clicking through those prompts caught a dangerous command 13.6 percent of the time.
Weekly digest
Every Monday, one email with whatever the archive added that week. Nothing lands in your inbox on a week the archive did not grow. One click to unsubscribe, any time.