AI is moving with force and speed across media, filling entire news portals, generating chart music and stepping directly into checkout carts. Content creators and publishers are left asking where their leverage lies, for revenue or for reach. A year ago Really Simple Licensing (RSL) promised an automated negotiator at the door, but thirteen months in, the machines aren't paying.

The number of articles Quartz published under one byline on a single day, 8 October, mostly rewrites of other outlets' reporting, around 20,000 words a day (Press Gazette). In April 2025 G/O Media sold Quartz to Redbrick, a Canadian software company, and the entire editorial staff was fired. Co-founder Zach Seward afterwards called it "a zombie brand" (Nieman Lab).

Why care? The brand is still there, the newsroom is not. The open question: will readers accept volume, or go where people do the reporting?

9 October 2026 · Kotaku

Within a couple of weeks, hundreds of AI-decompiled games and emulators appeared online. Kotaku writer Lewis Parker played browser versions of Halo, GTA: Vice City, Call of Duty: Black Ops and The Simpsons: Hit and Run. Some ran almost flawlessly. One Halo port had working online servers (Kotaku). Kotaku links the wave to Claude Opus 5.5.

AI can now read a game's compiled code and rebuild the source overnight, work that used to take skilled coders months. These games were built to resist copying. If they can be refactored this fast, the same goes for any protected content, from film and music to text and images.

12 October 2026 · Meta / Axios / Shopify

Meta released Muse Wonder on 12 October, the model that will soon run Muse, its personal AI agent. Muse is set up as a service: if users want, it books appointments or shops for them. Muse launched in the US on 8 September (Axios). In a nutshell the user benefit is that personal data, credit card details and the shopping history are saved, so this is about convenience. From the Meta perspective they are moving from advertising to being part of the deal, accumulating a shopper's history over time. This is much more powerful than advertising. They are not just sending people to a shop, they are present until the deal is done. Muse had 2.5 million downloads in the first two weeks (Sensor Tower via CNN, reported by eMarketer). Shopify signed a deal and introduced "Shop Pay" checkout on all its stores (Retail Insight Network). Amazon, though, blocked Muse, for now. Amazon says Muse does not identify itself to websites while browsing, and that it stores customer credentials in risky ways (OfficeChai). According to Meta a safety report is "coming soon".

8 October 2026 · CNN / OpenAI

Seven fake freelance journalists, built with ChatGPT, placed almost 100 articles in roughly a dozen online outlets, according to OpenAI's "Bogus Bylines" threat report (CNN). The operators prompted ChatGPT in Farsi to polish pitches and articles. Then they wrote fake comments reacting to the published pieces.

Status: the numbers are OpenAI's own. CNN says none of the outlets was a major mainstream one. Middle East Monitor removed the flagged pieces and is reviewing how it takes submissions. OpenAI says it is unclear whether the Iranian government is behind it.

29 September 2026 · Official Charts / EDM.com

"Bad Times" by Imael Angel entered the UK Singles Chart at #97 on 29 September (Songfacts). Imael Angel is Bernadett Gángó, from Hungary. She writes spiritual lyrics and turns them into full songs with AI tools. An earlier song of hers became "Movin' To The Sun" by the French DJ Hugel, a UK #2 (Official Charts). And in Oregon, a visually impaired guitarist runs a Christian rock band called "Shunned at a Funeral", which is fronted by three AI-generated women (KOBI5). Its short videos on YouTube look like a real band playing to big crowds (one example, careful it gets loud). And the visual quality will improve further, be sure about it. There might be an AI supergroup in just a few years, like it or not.

The situation now: the Recording Academy removed Hugel's track from Grammy consideration over AI concerns. Hugel is appealing and says about 95% of the release is human-made (EDM.com). The Academy allows AI material if there is enough human authorship. It will now be decided where the line is, but the perspective is that AI use will be normalised soon, first in dance, then in pop music. People will argue that AI is not that different from a synthesizer or sampler. Deezer said in July 2026 that it receives nearly 90,000 fully AI-generated tracks a day (Deezer).

Any owner of any form of content is currently on high alert and searching for ways to do business with AI engines. The perspective is that in just a few years agentic AI might be the new filter for many users on the internet. They might be on Facebook, but with a filter. They might have a starting screen where all their preferences, wishes and interests are stored. This means: they might increasingly not go to a specific website, not even to YouTube or TikTok, but have a highly individual channel feed. This is a problem, of course, for any platform or publisher.

What it is. RSL (Really Simple Licensing) was introduced only about one year ago, on 10 September 2025. Behind it are Eckart Walther, a co-creator of RSS, and Doug Leeds, former CEO of Ask.com and IAC Publishing (launch release). The organisation behind the standard is a nonprofit, the RSL Collective. The legal structure is that of a rights organisation. The main goal is to license content for many publishers at once, like ASCAP and BMI do for music. Membership is free. The Collective can make money by taking a cut only when publishers get paid (WAN-IFRA). From the article it was not clear how high that cut would be.

How it works. Deployment is simple. A website needs just one extra line in its robots.txt file, pointing to a licence file. The file says, per use, what is allowed, what is not, and at what price: free, attribution, subscription, pay per crawl, or pay each time the content is used in an AI answer (spec). Optionally, a crawler gets a licence token. Without one, the site can answer with "402 Payment Required" (crawler guide). By the way, here is one correction: in Issue #5 I wrote "rsl.txt". The spec defines a License line in robots.txt pointing to an XML file.

Where it stands. The spec is still version 1.0 from 10 December 2025 (press page). RSL says more than 1,500 organisations support it. That counts endorsements, not websites that use it (datalicenses.org). The only independent scan I found checked 42 sites. Two had an RSL licence. Thirty-one simply blocked AI crawlers (ScrapingBee). But here is the current problem: so far no AI company has publicly committed to honour RSL. And I found no reported payout. In April 2026 John Boyden of the RSL Collective put compensation for most publishers at "effectively zero" (WAN-IFRA, above), though in context this described the situation and the need for something like RSL.

Where the money moves: Instead of one strong standard, for now individual deals or blocking and charging are the approaches taken. For the AI engines this is the better way, instead of agreeing on terms with one publisher and quickly having demand from multiple content owners to get the same deal. For now, the bigger, stronger content or platform owners have their own strategies. For example, since 15 September, Cloudflare has blocked training and agent crawlers by default for new and free-plan sites (TechCrunch). On 30 September it put “Pay Per Use” into beta: the buyer offers a price and reports each use (Cloudflare). In another case Google pays about 100 publishers in its own pilot (#22).

The case against. Thomas Baekdal argues that paying per crawl is a one-time fee for content AI reuses endlessly (Baekdal). Plagiarism Today rated Cloudflare's approach stronger, because it actually blocks bots that don't pay (Plagiarism Today). No court or regulator has cited RSL yet.

What works. It is cheap. There are WordPress plugins, and Medium uses RSL to forbid training while allowing summaries with links (Medium). In Spain, the rights organisation CEDRO became RSL's exclusive partner (IFRRO).

One small fact I like. RSL presents itself as the heir to RSS. Reddit, one of its founding backers, will switch off RSS feeds of Reddit content on 13 November (TechCrunch). The reason: Reddit, with millions of user comments, has become a key source for AI training. So the company now aims to sell its data in direct deals instead. While confusing, this might be a very specific case. Reddit has bargaining power that smaller sites do not have.

For now I read RSL as a sign on the door. The approach is logical, fair and simple. But business is not about being fair, but about negotiating towards the lowest possible costs. So RSL states your terms in a form machines can read, and it costs minutes. Whether anyone reads the sign depends on who controls the door. Right now that is mostly Cloudflare. For now RSL is an experiment, but the current iteration might need either new laws or a change of the market to work.

Open questions:

Build, buy or borrow. Grzegorz Piechota of INMA has a great blog post worth your time where he lists the ways into a news brand using a simple grid: build, buy or borrow, crossed with awareness, visit, identity and payment (INMA). The simple grid might help many publishers to think about and decide what the future of monetisation might be. My impression: Arguing that content can be the main source of revenue was and is problematic. If you look at the history of newspapers and some magazines, they usually made money through packages and multiple services, but never solely through content. Content plus audience plus advertising plus door delivery was a very profitable formula (or package) for decades. Now, Piechota has some great examples of news brands extending their revenues through additions.

A fact-checker that got cheaper. Verso's free plugin "big if true" turns Claude into a fact-checker. Co-founder Patrick Swanson says one check went from 16 minutes and $9 to 6 minutes and $1.33 as the models changed, still catching about 9 in 10 known errors (Verso). Vendor-stated, and it runs on Claude, which I use.

Watermarks, two speeds. Google opened its SynthID checker for images, video and audio to everyone on 7 October (TechCrunch). OpenAI will watermark ChatGPT text, but only in the EU (TechCrunch). By OpenAI's own numbers, swapping 10% of words for synonyms drops detection from about 92% to 66% (OpenAI).

USA Today Co sues OpenAI. The publisher and its local papers filed in New York on 8 October, seeking more than $250 million. The same tracker lists Reach joining Google's AI pilot (#22) (Press Gazette).

TAUS Conference 2026 · 14 to 16 October 2026 · Vancouver · taus.net · Theme "intelligence-as-a-utility", right before LocWorld in the same city.

LocWorld56 · 19 to 21 October 2026 · Vancouver · locworld.com · The localisation industry's largest event.

Languages & The Media · 4 to 6 November 2026 · London · languages-media.com · Disclosure: I have a speaker slot.

Language Intelligence 2026 · 17 to 19 November 2026 · Vienna · eventbrite.at · Panel on why language models need structured knowledge, with APA, Infineon, Klartext AI and Intertranslations.

Yet another example where AI changes communities, here it is about math: on 6 October OpenAI published hundreds of AI-written proofs for about 370 open math problems in one go (Fortune). Mathematicians surprised, impressed and divided. NYU's Tristan Buckmaster told the New York Times it is unclear how much the human mathematicians steered the results, and now humans have to check how good the results really are. At this volume that will take a lot of time and effort (NYT, paid).

I am Mirko Lorenz. I work on language technology projects at Deutsche Welle in Germany.

Three projects you will hear about in this newsletter:

AI use: Full disclosure, I do use Claude (Anthropic) to research and edit this newsletter, with prompts I have refined many times. My goal is to find out where AI is reliable and where the hallucinations come in. Before publishing I check all facts and links.

Error log: To track the issues of using AI I maintain an open Google Doc where I collect the small and big problems that a tool like Claude introduces to editorial work. The idea is to get a better understanding of where AI is good and where it is not. Read it here.

babylon-newsletter.com · 7,000 languages. AI works for 20.