We've finally launched our updated website. Still some work to do, with more coming soon... infonomic.io

Three years into the AI copyright fight, the headline question is still open: no court has settled whether training on copyrighted work is legal. But the rulings so far point somewhere more specific. Courts are not really deciding whether machines may learn from text. They are asking two narrower questions. Where did the material come from, and can anyone prove the output harmed the market for the original? Almost everything decided so far turns on one of those two. Provenance first. On 20 July 2026 a federal court gave final approval to the $1.5 billion settlement in Bartz v. Anthropic, roughly $3,000 for each of about 500,000 books, plus an obligation to destroy the pirated files. The underlying ruling had split the question cleanly: training on lawfully acquired books was fair use, keeping a permanent library of pirated copies was not. Anthropic was penalised for how it got the material, not for training on it. The release covers acquisition and copying up to August 2025 and nothing else, so output claims survive. That sequence has been said out loud. Speaking to an AI class at Stanford in 2024, the former Google chief executive Eric Schmidt described the play for an AI startup: tell the model to build your competitor and "steal all the music", get it in front of people, and "hire a whole bunch of lawyers to go clean the mess up" if it works. He asked for the video to be taken down and said later that he had not meant it literally. It remains a fair description of the incentive. Clearing rights up front costs money now, while a ruling years away is uncertain and discounted. Bartz is the first time that bill has been presented, and whether $1.5 billion is large enough to change the calculation is an open question. Then harm, which is where most cases have actually turned. In Kadrey v. Meta the authors lost, and the judge said plainly that they lost because they had not proved market harm, not because training is lawful. He also flagged a dilution theory a better-argued case could win. Thomson Reuters v. Ross failed for the mirror-image reason: the copying produced a product that competed directly with the original. Harm is the hinge, and it is hard to evidence. Neither question is settled at the level that counts: no appeal court has ruled on any of it. The first to hear argument was the Third Circuit on 11 June 2026, in Ross, and the panel spent most of its time on exactly these points, whether the use was transformative and whether it harmed the market, including a licensing market for training data. No decision has issued. NYT v. OpenAI has an order to hand over 20 million anonymised ChatGPT logs, summary judgment pending, and no trial date. Getty's UK case failed on its central copyright theory and is under appeal. One filing from last month matters more to this audience than any of the above. On 14 August a group of textbook authors sued OpenAI, following a parallel suit against Meta in July, and their harm argument is built around how academic work is actually bought. Textbook adoption is an institutional decision rather than a consumer one, so a substitute does not have to be as good as the book. It only has to be good enough for the committee that chooses it. That is a far easier harm to demonstrate than a lost novel sale. To summarise. Whether AI training is lawful in the abstract remains undecided, and will stay that way until an appeal court speaks. What the cases turn on is provenance and provable harm. For researchers, NGOs and archives that has a practical edge: clean records of what you hold and where it came from are becoming legally load-bearing. And every plaintiff named here is a large publisher or a well-organised class, not a small archive. Next: the incentive trap. #AIcopyright #ScholComm

Back from our first #IFLA #WLIC #WLIC2026. A great event with many excellent sessions, talks, posters as well as incredibly kind and helpful delegates. Hoping we can go to #WLIC2027

The first two posts in this series described the problem: original work ingested without consent or credit, and AI answers satisfying the reader in place of the source. The fair question is what is being built in response. The honest answer: more than you might think, less than you would hope. The fastest mover is infrastructure. Since July 2025 Cloudflare blocks known AI crawlers by default on new sites. Its Pay Per Crawl marketplace lets a publisher allow, charge, or block each crawler, and by Cloudflare's own April 2026 figures the network returns more than a billion "402 Payment Required" responses to AI bots a day. A companion Content Signals Policy adds machine-readable lines to robots.txt separating three uses: search, ai-input, ai-train. "Do not train on this" is now a default a small publisher gets for free. Then licensing. Really Simple Licensing (RSL), from the people who built RSS, became an official standard in December 2025. It turns robots.txt's yes/no into machine-readable terms, including pay-per-crawl and pay-per-inference. The RSL Collective bundles the long tail (the small publishers who cannot negotiate alone) into something an AI company will transact with. 1,500+ organisations have signed on, from the AP to Yahoo. Underneath, the IETF's AIPREF group is standardising one shared "Content-Usage" signal to replace today's patchwork. Real, moving, not finished. In parallel, C2PA Content Credentials attach cryptographic provenance to a file, and the EU AI Act's machine-readable marking duties take effect 2 August 2026. So the machinery is arriving: a way to say no, a way to set terms, a way to mark provenance. Two caveats. Published is not honoured: a signal only works if AI firms respect it. And none of this brings back the visit. A signal that blocks or prices a crawler is a defence, not discovery. The answer engine still satisfies the reader without a click, and attribution inside an answer is not the same as being read. For mission-driven publishers the standards are necessary but not sufficient. Worth adopting (mostly free, fast becoming table stakes), but they protect the boundary, not the relevance. Next: what the courts have actually settled. #AIcopyright #ScholComm

For twenty-five years the web ran on an implicit deal. Publishers let crawlers index their content freely. Search engines sent readers back as traffic. Nobody wrote it down. AI answer engines have ended the deal unilaterally. Similarweb measured zero-click searches (queries that end without a single click to any website) at 69% by May 2025. By 2026 the figure is above 80%. A Pew Research Center study of 68,000 real queries found click-through rates fall by half when an AI summary appears: 8% versus 15% without one. Only 1% of users click the citations listed inside the summary. Ahrefs measured the click-through rate for content ranked first on informational queries, the kind where knowledge matters most, and found it had collapsed from 7.6% to 1.6%. A 79% decline, for content ranked first. On the other side: Cloudflare data puts Anthropic's crawlers at roughly 70,000 page requests per referral visit sent. The AI system takes at scale; it returns almost nothing. For ad-funded publishers this is a revenue crisis. Some have closed entirely. "The extinction-level event is already here," said Helen Havlak, publisher of The Verge. For researchers, NGOs and archives, the threat is different and in some ways sharper. Their revenue doesn't primarily depend on traffic. Their credibility does. Citation and attribution are the currency that academic and mission-driven work runs on — they justify the grant, validate the methodology, establish the institution. An AI answer that draws on research without naming it, or naming it in citations 99% of readers never click, doesn't just lose the visit. It breaks the attribution chain that makes the work legible as knowledge rather than noise. A paraphrase without a name attached makes an institution disappear in plain sight, even while its work is actively in use. The question is not only: how do we recover the traffic? It is: how does original work establish, in the age of AI interfaces, that it exists, is trustworthy, and came from somewhere specific? Those are not SEO questions. They are questions about what it means to be a canonical source. Next: what the standards bodies are building in response. #AIcopyright #ScholComm

In the AI copyright fight, there are two kinds of rights-holders: those big enough to sue, and those big enough to sign. The New York Times, Disney and Getty are litigating. News Corp signed with OpenAI for a reported $250M over five years. Reddit licenses its corpus to Google for a reported $60M a year. Content, it turns out, has a price for those with the scale to negotiate. Below that line sits almost everyone who actually creates original work: independent publishers, individual academics, NGOs, cultural archives. Too small to sue. Too small to license individually. Ingested all the same: no consent, no credit, no cheque. And the grievance isn't only about money. When Taylor & Francis licensed academic content to Microsoft for a reported $10M, the authors were neither consulted nor compensated, and could not opt out. The backlash from researchers was less about the cheque than the consent: being used without permission, mis-cited, or stripped of context. For researchers, NGOs and archives, credibility is the asset. The real risk of the AI era isn't lost ad revenue. It's invisibility and mis-attribution, as AI answers quietly replace the visit to the source. A paraphrase with no name attached makes a body of work disappear in plain sight. First in a series on attribution, consent and the economics of original content in the AI era: what's settled, what isn't, and what small, mission-driven publishers can practically do about it. #AIcopyright #ScholComm

TIL: content negotiation at a single URL — when your CDN doesn't honor Vary by header — looks like this 👇 That's the Next.js RSC flight protocol getting dumped straight onto the visitor's screen. Ouch. The App Router serves the HTML doc and the text/x-component RSC payload from the same URL, split only by the RSC request header. Next sets Vary: rsc correctly… but Cloudflare ignores Vary on everything except Accept-Encoding — so both variants collapse into one cache entry, and whichever lands first gets served to everyone. Fix: a Cache Rule to bypass cache on RSC requests. (Mind the ordering — Cache Rules are "last match wins," the opposite of WAF rules.) Looking at you, Cloudflare, Next.js + RSC 👀 #nextjs #cloudflare #rsc #webdev #caching

We still believe in the open web — and in curated public knowledge. Because whether you love AI, hate it, or feel uncertain about it… AI runs on content. infonomic.io/platforms/colle… #archives #Collections #knowledge #content

We've launched FORRU-CMU - a great project for a great cause... forru.org/ #forestry #climate #environmental

Migrating from Drupal to Payload CMS: A long-ish post that describes how - and more importantly, why, we migrated from Drupal to Payload CMS. infonomic.io/blog/migrating-…