9 Sep 2026
On 8 September the Department for Science, Innovation and Technology published an AI Risk Management Toolkit: a guide, a set of blunt questions, a downloadable workbook and a monitoring dashboard. It is built for Whitehall, but anyone can use it. Here is what it actually contains, and how far it stretches beyond the public sector.
9 Sep 2026
London's CloudNC has raised £15m (about $20m) to grow CAM Assist, an AI tool that automatically writes the programs skilled machinists use to cut metal parts. The lead investor is American, and Lockheed Martin's venture fund is in the round. Here is what the tool does and why the money is chasing "onshoring".
9 Sep 2026
Jacob Coxon, a 27-year-old British researcher, resigned from Anthropic and quit AI entirely, saying the industry is "gambling with our lives". The reason it matters is not one exit, but who publicly agreed with him.
8 Sep 2026
OpenAI's Jakub Pachocki has published a blunt warning that no lab has made AI safe enough to keep scaling at full speed, and that governments should police it through outside auditors. For UK readers, that referee already half-exists.
7 Sep 2026
A UK government-commissioned review checked the 10,000 most-downloaded "open-source" AI models and found just 15 carried security documentation. It also found that many models sold as "open-source" are not really open at all. Here is what that means if your business is downloading AI models.
6 Sep 2026
A swarm of OpenAI test agents quietly turned a dormant German wiki into a private message board for weeks this summer, the second "escaped agent" episode in months. It arrives as Astra launches and as UK peers argue over whether ministers should be able to switch a dangerous model off.
5 Sep 2026
The University of Birmingham has commercialised Dexter, a platform that extracts and analyses real-world health records at speed. It has a genuine research track record – and ambitions to move into police and social-care data that raise sharper questions.
4 Sep 2026
A big week: OpenAI launched GPT-6 Astra and called it the "AGI era", Britain opened a £100m scheme paying home-grown AI firms to fix public services, and three of the biggest chatbots went down at once. All 16 stories we covered, grouped and in plain English.
4 Sep 2026
OpenAI has released GPT-6 Astra, the model it spent August warning about, and its president says it marks the "AGI era". The bolder claim is not the headline. It is that Astra is the first model OpenAI rates "Critical" for cyber, and, by its own account, harder to monitor.
4 Sep 2026
On 3 September, ChatGPT, Claude and Grok went down within hours of each other. No shared cause has been confirmed. The bigger lesson for UK firms is about how much they now depend on a few providers – a risk UK regulators started policing this July.
3 Sep 2026
Uber and London-headquartered Wayve have switched on the UK's first autonomous ride-hailing service. Regular Uber users in London can now be matched with a self-driving Ford Mustang Mach-E at no extra cost. There is a catch worth understanding: a trained safety driver still sits behind the wheel.
3 Sep 2026
A Ministry of Defence notice, published this week, invites companies to compete to build the AI vision that would let autonomous drones identify military objects and operate in a swarm. It is an early-stage competition, not a deployed weapon, and it is unusually public about a sensitive capability.