The global network under everything
Undersea cables, DNS and the web. How billions of devices agree on where to send each byte.
Short visual breakdowns of how the internet works. Open one to watch it right here.
The newest Internet stories from the daily feed, explained in plain language.
The Verge
Bluesky's Attie tool started as a way to build custom feeds with AI prompts. It now generates small social apps, such as games, polls and dashboards, each published as its own website with a link and credit for its maker. Bluesky's CEO calls it vibe coding for your community, and a simple poll takes a few minutes to create.
CNET
YouTube Music is adding labels to songs it identifies as created primarily with generative AI, using its own detection tools and information from music partners; the tags will come later to the main YouTube app. Platforms are reacting to a surge of AI tracks: Deezer reported that AI music made up 44 percent of new deliveries, Apple Music plans labels by year's end, and Spotify offers AI Personas.
The Decoder
arXiv is a free online library where researchers post papers before peer review, and volunteer moderators screen the flood. The site says monthly submissions passed 40,000 in September, double the level two years earlier and more than six times higher in its AI category, so from October 1 each person may post only two papers a month; rejected papers still count.
The Verge
ICANN is taking applications for new top-level domains, and the 1,615 submissions are full of AI branding. Ten applicants, including Meta and OpenAI, want.agent; seven want.agi; OpenAI also applied for.chatgpt and.gpt, while Anthropic wants.claude. The suffixes are only proposals so far, with nothing official before next year, but they show how much the industry expects AI agents to shape the web.
The Next Web
Personal agent protocols set rules for how an AI assistant signs into shopping and other sites and what it may do there. The Personal Agent Protocol from Sierra and Meta uses OAuth sessions: an agent can browse as a guest, then a signed-in customer chooses between read-only access and write access, and each company decides whether agents use its website, its APIs or its own bot.
The Verge
AI agents gather information by making automated requests to websites, and Wikimedia says it traced activity from OpenAI agents on its platforms: test edits in wiki sandboxes, attempts to use its note-taking tool as a data proxy, and millions of API requests that the foundation says may have helped cause a partial outage in May. OpenAI says it is reviewing the findings and has not confirmed any link to the outage.
The Register
arXiv now limits each submitter to two new papers per month and three active submissions, after AI-assisted low-quality uploads overwhelmed its moderators. The preprint server took in 40,363 submissions in September, double the number two years earlier, and rejected papers still count against the monthly cap.
The Verge
Cloudflare runs a global network that sits between many websites and their visitors, so bots hit Cloudflare before they reach those pages. That position lets site owners block AI crawlers and agents, allow them, or let in only the ones that pay. With Cloudflare saying bots now make up more than half of internet traffic, CEO Matthew Prince talks through how that control works and what it means for the web.
