From editorial guidelines to architecture: Twenty newsroom leaders from 13 countries explain how AI governance is moving beyond policy documents and into the infrastructure itself. (Reuters Institute)

There is a new genre: writers describing their struggle to produce words, something AI can never experience. It’s really boring. Monocle editor Andrew Tuck at least compares AI-assisted writing to taking Viagra, and notes in passing that he has human assistants to clean up his copy.

The Dissent is a San Francisco news site written and edited by eight AI agents. Either the bleak future of local journalism or a very elaborate bit. Possibly both. Mission Local reports that it was built by a BlackRock portfolio specialist in his spare time and that it uses the work of human journalists.

How AI text watermarks hide inside word choices, not in the characters themselves: a visual, step-by-step explainer. Covers also how to erase marker. (James Padolsey)

Declaude: Paste in your AI slop, get back plain prose, without the hidden watermark prescribed by the EU. Each IP address gets 50 requests a day, so hurry before your colleagues discover it.

Swiss public broadcaster SRG SSR makes several thousand hours of radio content available to train Apertus, Switzerland’s open LLM. (via Brecht Declercq)

Take a 28-question quiz to find out where you land on AI. Thirty archetypes, each with a patron saint ranging from Emily Bender to Ray Kurzweil. I got “The Optimist” instead of “The Journalist”, which feels about right.

Waiting for the next Claude or GPT drop? OutYet.ai tracks upcoming model releases.

Interface, context, harness, model, governance, and emergence: You cannot do good AI product work without understanding what happens below the chat box. (Emily Campbell)

David Caswell argues legacy newsrooms will not fundamentally transform for AI but will instead optimize existing workflows until they hit a wall built from old assumptions about articles and human-in-the-loop processes. Long interview, worth it. (Mirko Lorenz, Babylon)

AI models’ values are more secular and more liberal than those of almost any human population on Earth – unless they’re made in China, in which case they’ll nudge you to reconcile with your meddling in-laws for the sake of family harmony. The Economist ran 25 frontier models through the World Values Survey.

Europe 2031: Speculative fiction about what Silicon Valley sees coming and Europe doesn’t. The story runs from DeepSeek to a 2031 where Europe has largely lost the ability to chart its own course.