Students edition
Utrecht gives students their own AI chat, AI bosses are hitting the brakes themselves, and a quarter of all parliamentary questions turn out to be partly written by AI.
Utrecht gives students their own AI chat, and that changes the privacy question
Free access to several language models, processed within the EU and not used to train models.
Since 9 September, all students and staff at Utrecht University can work with UU AI Chat, the university's own platform for generative AI. It gives you free and equal access to different language models. The environment meets the university's privacy and security requirements. Your data is processed within the EU. It is not used to train external AI models. The university calls this a step towards digital autonomy: more control over where your data goes. The platform was built for education first, but it can also help with research and other work. Ask your own programme whether such an environment exists or is coming. A platform like this is, at its core, a shell around existing language models. You type your question into the university's environment. It sends your question to a model and passes the answer back to you. The difference is not in the model, but in the agreements around it: where the data is processed and what may be done with it. That's why you can discuss study material here that you'd better not paste into a public chatbot. Watch out for two limits. Safe processing is not the same as permission from your programme: the rules on AI in assignments and theses do not change because of this. What is and isn't allowed for a submission is written in your study guide. And the model is still the same model, with the same tendency to confidently say wrong things. So keep checking facts against your study material. Ask your programme whether they also offer such an environment, and whether you're allowed to use it for assignments.
Background for subscribersThis edition touches you in several ways. Utrecht gives you free, secure access to AI chat, but that changes nothing about your programme's rules on AI in assignments. The Gemini incident shows you need to think carefully before giving an AI tool access to your files or email: that's hard to undo. The parliamentary questions affair shows that hiding AI use weighs more heavily than the use itself, exactly as with your own accountability statements. The legal test and the IND ruling both show that AI is good at naming sources but struggles with real application, and that is exactly what you get graded on. And the ChatGPT study warns: the easiest way to use it feels the most educational, but delivers the least.
AI bosses are asking for a brake themselves, and want outside oversight
Over the weekend of 12 and 13 September, Anthropic boss Dario Amodei published the essay 'We Must Pace the Frontier'. His argument: the sector must slow the pace of new capabilities. He points to self-improving systems and risks such as cyberattacks and bioterrorism. He proposes three steps. The first is independent overseers who get permanent access to the models made by AI companies. Sam Altman (OpenAI), Demis Hassabis (Google DeepMind) and Elon Musk (xAI) publicly backed the idea, which is notable, since these are rivals who normally race to build the fastest model. Anthropic already took a first step: Accenture becomes the first external reviewer with employee-level access, to check whether Anthropic keeps its own safety promises. Anthropic is calling on the rest of the sector to do the same. The company also reported that Claude itself already directs 26 percent of the research into Claude's successor, under human supervision. 'Independent oversight' sounds vague, but the detail is concrete: Accenture gets access at the level of an employee. That's not a report afterwards, but watching the process itself as it happens. This is needed because you can't tell from the outside whether a company follows its own safety rules. The weak spot is that joining in is voluntary: if you're the only one slowing down, you lose ground to a competitor who doesn't. There's something else in this story: Claude already directs a quarter of the research into its own successor. That's exactly the kind of self-improvement Amodei calls a risk. So the company is warning about something it is doing itself. Take calls to slow down seriously, but don't count on them: choose your study tools based on what they can do now.
Background for subscribersThis is a taster.
Subscribers read the whole Students edition: every story, each with its background in plain language.
Subscribe →Not ready for that? Read the free letter first →
AI and tech, made simple.
This is the Students edition. Subscribers get the background to every story, in plain language.
Subscribe
inklaretaal · Amsterdam · © 2026